On Mon, Aug 18, 2025 at 2:29 PM F6BVP [off-list ref] wrote:

I agree that it must be the same bug and mkiss module is involved in
both cases although the environment is quite different.
I am using ROSE/FPAC nodes on different machines for AX25 messages
routing with LinFBB BBS.
Nowadays I do not have radio anymore and all are interconnected via
Internet using IP over AX25 encapsulation with ax25ipd (UDP ports).

I am running two RaspBerry Pi 3B+ with RaspiOS 64Bit and kernel 6.12.14.
AX25 configuration is performed via kissattach to create ax0 device.
ROSE / FPAC suite of applications manage ROSE, NetRom and AX25 protocols
for communications. FBB BBS forwards via rose0 port and TCP port 23
(telnet).

I do not observe any issue on those RasPiOS systems.

Another mini PC with Ubuntu 24-04 LTS and kernel 6-14.0-27-generic is
configured identiquely with FPAC/ROSE node and have absolutely no issues
with mkiss, ROSE or NetRom.

A few years ago I had been quite active on debugging ROSE module. As I
wanted to restart AX25 debugging I installed Linux-6.15.10 stable
kernel. This was the beginning of my kernel panic hunting...

My strategy is to find the most recent kernel that do not have any issue
with mkiss and progressively add AX25 patches in order to find the
guilty instruction. I will use a buch of printk in order to localize the
wrong code. We will see if it works.

Bernard,

    Very good. A caveat is that the issue seems to be the bug
manifests itself in the `skbuff` infrastructure, independent of the
specific AX.25/NETROM/ROSE code: it may be that some other change
elsewhere in the kernel failed made a change that was incompatible
with AX.25 that gave rise to this bug.

    I've found the oops to be very reproducible. Given that you seem
to have a known working kernel version, you may get more mileage out
of using `git bisect` to narrow things down to a specific failing
commit, instead of trying to forward-apply AX.25-specific commits.

        - Dan C.

Le 18/08/2025 à 18:30, Dan Cross a écrit :

quoted

On Mon, Aug 18, 2025 at 6:02 AM Bernard Pidoux [off-list ref] wrote:

quoted

Hi,

I captured a screen picture of kernel panic in linux-6.16.0 that
displays [mkiss]. See included picture.

Hi Bernard,

     This is the same issue that I and a few other folks have run into.
Please see the analysis in
https://lore.kernel.org/linux-hams/CAEoi9W4FGoEv+2FUKs7zc=XoLuwhhLY8f8t_xQ6MgTJyzQPxXA@mail.gmail.com/#R (local)

     There, I traced the issue far enough to see that it comes from
`sbk->dev` being NULL on these connections. I haven't had time to look
further into why that is, or what changed that made that the case. I
now think that this occurs on the _first_ of the two loops I
mentioned, not the second, however.

         - Dan C.

(Aside: I'm pretty sure that `linux-hams@vger.kernel.org` is not a
Debian-specific list.)