mvebu: update to kernel 6.1

[lynxis]

Describe the bug

To collect all related bugs and have a discussion of the missing parts for kernel 6.1 for the mvebu target.
The first steps have been made #12938 to add 6.1 as testing kernel.

OpenWrt version

master

OpenWrt target/subtarget

mvebu

Device

all

Image kind

Self-built image

Steps to reproduce

No response

Actual behaviour

No response

Expected behaviour

No response

Additional info

No response

Diffconfig

No response

Terms

• I am reporting an issue for OpenWrt, not an unsupported fork.

[stklcode]

For reference, my key takeaways from the initial PR:

Running on a CZ.NIC Turris Omnia (cortexa9) with dual MT7915 radios (mt7915e driver) for about a month (currently on 6.1.42) with quite a few packages and configs
=> no obvious regressions from 5.15 so far

Also running on a Linksys WRT1900ACS (cortexa9) as testing device.
=> Marvell WiFi drivers (mwlwifi) do not compile without additional work on the driver. See #12940 or the equivalent upstream proposal (kaloz/mwlwifi#415).

[lynxis]

I'm running it on cortex a72 / clearfog cn9130 base (with additional patches).

[tiagogaspar8]

I'm not being able to compile mwlwifi on kernel 6.1:

/mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714/hif/pcie/pcie.c:1296:9: error: unknown type name 'mm_segment_t'
 1296 |         mm_segment_t oldfs;
      |         ^~~~~~~~~~~~
/mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714/hif/pcie/pcie.c:1305:17: error: implicit declaration of function 'force_uaccess_begin'; did you mean 'user_access_begin'? [-Werror=implicit-function-declaration]
 1305 |         oldfs = force_uaccess_begin();
      |                 ^~~~~~~~~~~~~~~~~~~
      |                 user_access_begin
/mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714/hif/pcie/pcie.c:1329:9: error: implicit declaration of function 'force_uaccess_end'; did you mean 'file_accessed'? [-Werror=implicit-function-declaration]
 1329 |         force_uaccess_end(oldfs);
      |         ^~~~~~~~~~~~~~~~~
      |         file_accessed
/mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714/hif/pcie/pcie.c: In function 'pcie_probe':
/mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714/hif/pcie/pcie.c:1556:14: error: implicit declaration of function 'pci_set_dma_mask' [-Werror=implicit-function-declaration]
 1556 |         rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
      |              ^~~~~~~~~~~~~~~~
cc1: some warnings being treated as errors
make[5]: *** [scripts/Makefile.build:250: /mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714/hif/pcie/pcie.o] Error 1
make[4]: *** [Makefile:2012: /mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714] Error 2
make[4]: Leaving directory '/mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/linux-6.1.42'
make[3]: *** [Makefile:103: /mnt/openwrt/ss_mod/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/linux-mvebu_cortexa9/mwlwifi-2023-04-29-6a436714/.built] Error 2
make[3]: Leaving directory '/mnt/openwrt/ss_mod/package/kernel/mwlwifi'
time: package/kernel/mwlwifi/compile#3.17#0.48#3.71
    ERROR: package/kernel/mwlwifi failed to build.
make[2]: *** [package/Makefile:120: package/kernel/mwlwifi/compile] Error 1
make[2]: Leaving directory '/mnt/openwrt/ss_mod'
make[1]: *** [package/Makefile:114: /mnt/openwrt/ss_mod/staging_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/stamp/.package_compile] Error 2
make[1]: Leaving directory '/mnt/openwrt/ss_mod'
make: *** [/mnt/openwrt/ss_mod/include/toplevel.mk:232: world] Error 2

Has anyone seen this?

[stklcode]

Yes, the mwlwifi drivers are not ready yet. See #12940

(affects all targets with Marvell WiFi chips, i.e. WRT1900* and WRT32*)

Update: patches are now part of the main branch

[tiagogaspar8]

Would you believe me if I told you that I actually searched all over the place for this? 🤦
These devices really are a pity...

[PalebloodSky]

Been testing kernel 6.1 with my WRT32X. Mvebu hasn't worked this well since the OpenWrt 19.07/kernel 4.14 era. VLAN leaks are fixed (between removing that broken hack and new 6.1 fixes), the new mwlwifi driver works better with latency fixed (6.1), SQM performance is excellent (500Mbit down with SQM cake is no problem), DSA performance is great, rock solid stability, etc.

Only thing broken is software flow offloading. But that has been since DSA I think, not a big deal since most people use SQM.

Confirming mvebu is working great on kernel 6.1. Please switch mvebu to 6.1, before 23.05 release would be ideal. Thanks to the devs for your efforts.

[stklcode]

23.05 will be on 5.15 only and IIRC the VLAN leak is already fixed.
It’s in RC2 state already, so highly unlikely a yet experimental kernel update will be backported or anybody even has plans to maintain two kernels in a stable release.

6.1 is target for the next generation, likely 24.xx.

But thanks for your testing and feedback. Sound promising 😄

[PedroM4]

what about the igmp issues that are fixed by removing the ARM-dts-turris-omnia-Add-switch-port-6-node patch? @stklcode did you have time to look into it?

[PalebloodSky]

@stklcode Yea VLAN is mostly fixed in 5.15, but not entirely. 6.1 fixes the remaining broadcast/multicast issue as @champtar confirmed here #12878. I confirmed from my tests as well. Nevertheless thanks for confirming the process it's appreciated.

[M95D]

OpenWrt r24154 with kernel v6.1.57 on mvebu/cortex-a9, Linksys WRT-1900ACv1. I built the kernel with io-uring and there is a problem with samba's uring VFS. I can't access any shares. Samba log:
make_connection_snum: SMB_VFS_CONNECT for service 'Archive' at '/mnt/Archive' failed: Function not implemented
If I remove /usr/lib/samba/vfs/io_uring.so and restart samba, everything starts working again.

Can anyone confirm?

[trippleflux]

@M95D
Please retry re-compiling Samba4 with my fixes posted in here. Would it work for your Samba4.