Hi Andre, Martin,
On 13.01.2025 14:17, Andre Heider wrote:
On 13/01/2025 9:17 am, Martin Schiller wrote:
On 2025-01-12 15:09, Sergey Ryazanov wrote:
atm_qos struct should be the same both for user and kernel spaces. Via
the __SO_ENCODE() macro it is used to define the SO_ATMQOS socket IOC.
During the VRX518 support introduction, the atm_trafprm sturct nested
into the atm_qos stucture was update with newer fields that are
referenced by the ATM TC layer of the VRX518 TC driver. These new fields
are intended to communicate information for extra traffic classes
supported by the driver. But we are still using vanilla kernel headers
to build the toolchain. Due to the atm.h header incoherency br2684ctl
from linux-atm tools is incapable to configure the ATM bridge netdev:
br2684ctl: Interface "dsl0" created sucessfully
br2684ctl: Communicating over ATM 0.1.2, encapsulation: LLC
br2684ctl: setsockopt SO_ATMQOS 22 <-- EINVAL errno
br2684ctl: Fatal: failed to connect on socket; File descriptor in
bad state
Hmmm... I have fixed this problem in the past by including the
"user_headers". See commit e53ed6af8744 (“linux-atm: use target specific
kernel headers”) [1]. Felix suggested this to me in [2].
Does this no longer work?
Good question, I didn't know about that patch.
Me too. As it was mentioned in cover letter, I've faced the ATM bridge
issue on the latest stable release - 23.05.5. And the e53ed6af87
("linux-atm: use target specific kernel headers") commit was introduced
after the branching and was never backported:
$ git log --oneline -n3 master package/network/utils/linux-atm/
3e024022c3 linux-atm: fix build with GCC 14
e53ed6af87 linux-atm: use target specific kernel headers
9aa3d5b345 linux-atm: Include linux/sockios.h for SIOCGSTAMP
$ git log --oneline -n2 openwrt-23.05 package/network/utils/linux-atm/
9aa3d5b345 linux-atm: Include linux/sockios.h for SIOCGSTAMP
806354ab53 linux-atm: Fix compile warning
I developed the extra fields disabling patch for the stable release and
then rebased. And then in V3 extended it to drop the Lantiq's ATM hacks
entirely.
BTW, I checked the Martin's fix in the latest master just now. It works
and picks the correct atm.h from the patched kernel.
At least on ipq40xx there's nothing writing the custom classes to qos-
>txtp.traffic_class, it's all just code that reads and checks for
those, so it's basically useless and dead code. This removal here allows
us to drop the atm.h patch.
I checked the git history, it looks like since 2007 only Lantiq's legacy
driver (now ltq-atm) and various compat patches (now
MIPS-lantiq-add-atm-hack.patch) are referencing
ATM_VBR_NRT/ATM_VBR_RT/ATM_UBR_PLUS/ATM_GFR.
We have two options now. (a) Keep the Lantiq's hack in the tree for both
platform and keep the include path amend in linux-atm. (b) proceed with
this patch and probably backport it to v23 and v24. And with a follow up
patch remove the corresponding ATM hacks from Lantiq and eventually from
linux-atm.
If that's the case too for lantiq, we could apply the same logic there
and make linux-atm shared again?
Yep. I like the plan 'b' for cleaning the tree from vendor's hacks, stop
spreading them to other targets and simplifying maintenance.
In any case, this set still works on my ipq40xx device and vdsl seems to
work just fine!
Have you gave a change to this series, Andre?
--
Sergey
_______________________________________________
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel
