[Ethtool]: Fixing Build Failure on ARM Platform

[vivekrnv]

Signed-off-by: Vivek Reddy Karri vkarri@nvidia.com

Why I did it

Compiling ethtool from source is causing ethtool unit tests to fail on ARM Platforms.

These tests are failing: (By default netlink-interface is enabled while compiling ethtool)
Link: (Test File Link)

FAIL: test-cmdline
==================

E: ethtool 16_char_devname! returns 1
E: ethtool 
127_char_devname0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcde returns 1
E: ethtool --change devname xcvr external returns 0
E: ethtool --change devname speed 100 duplex half port tp autoneg on advertise 0x1 phyad 1 xcvr external wol p sopass 01:23:45:67:89:ab msglvl 1 returns 0

FAIL test-cmdline (exit status: 1)

Tested this on Local ARM Emulated Container:

(Docker Container Emulating ARM)
vkarri@3a03c70eed35:/tmp/ethtool$ ./ethtool 16_char_devname!
netlink interface initialization failed, device name longer than 15 not supported
vkarri@3a03c70eed35:/tmp/ethtool$ echo $?
1 (Expected 0)
vkarri@3a03c70eed35:~/ethtool$ ./ethtool 16_char_devnameee
netlink interface initialization failed, device name longer than 15 not supported

Checked for dependencies: (all are present)
vkarri@3a03c70eed35:~/ethtool$ apt-cache policy libmnl0
libmnl0:
  Installed: 1.0.4-2
  Candidate: 1.0.4-2
  Version table:
 *** 1.0.4-2 500
        500 http://deb.debian.org/debian buster/main armhf Packages
        500 http://packages.trafficmanager.net/debian/debian buster/main armhf Packages
        100 /var/lib/dpkg/status

vkarri@3a03c70eed35:~/ethtool$ apt-cache policy libc6  
libc6:
  Installed: 2.28-10
  Candidate: 2.28-10
  Version table:
 *** 2.28-10 500
        500 http://deb.debian.org/debian buster/main armhf Packages
        500 http://packages.trafficmanager.net/debian/debian buster/main armhf Packages
        100 /var/lib/dpkg/status

How I did it

Disabled netlink-interface for ethtool.

Even though Netlink is not available, it doesn't seem to impact what ethtool was supposed to do. In fact the older version which was in use before this PR #5725 did not have netlink support and everything seemed to work well

Article on Netlink-Support for ethtool: https://lwn.net/Articles/783633/

How to verify it

Build Passed on Local Slave ARM Emulated Container and on Mellanox x86 platforms

Which release branch to backport (provide reason below if selected)

• 201811
• 201911
• 202006
• 202012

Build Log

ethtool_5.9-1_armhf.deb.log.txt

A picture of a cute animal (not mandatory but encouraged)

[vivekrnv]

@shlomibitton Can you review this?

[liat-grozovik]

@shlomibitton please review the fix as it is on top of your PR.
@vivekredddo we need this on top of the reverted PR or this one replaces + fix the original one?

[vivekrnv]

@vivekredddo we need this on top of the reverted PR or this one replaces + fix the original one?

@liat-grozovik This change is needed on top of the original PR #5725.

For backporting to 202012 though, this can be followed. This commit (#5725 Revert Commit) can be reverted back and #5725 Fix can then be backported to 202012

[shlomibitton]

@shlomibitton Can you review this?

LGTM, but I did a test to check this issue and used a real system based on ARM CPU x64, the build for the tool succeeded without this fix.
@lguohan @qiluo-msft could you check this please? maybe it is a build environment issue?

[vivekrnv]

@shlomibitton Can you review this?

LGTM, but I did a test to check this issue and used a real system based on ARM CPU x64, the build for the tool succeeded without this fix.
@lguohan @qiluo-msft could you check this please? maybe it is a build environment issue?

Hi @shlomibitton,

Not sure if these builds ran on real ARM server or an emulated one, the ARM builds were failing.
Azure Pipelines Run

[lguohan]

the build is running on qemu emulation. are you saying the test is working fine on real arm cpu?

in this, should we disable the unit test, instead of disable the configuration option.

[vivekrnv]

@lguohan
Done, disabled tests instead of the netlink configuration option

[lguohan]

NO NEED TO IT HERE.

you can refer this example.

rules/sairedis.mk:$(LIBSAIREDIS)_DEB_BUILD_OPTIONS = nocheck

it will disable check

[vivekrnv]

Thanks for the example, @lguohan

[yxieca]

@vivekreddynv this change cannot be cherry-picked to 202012 branch cleanly, please create a separate PR for 202012 branch.

[vivekrnv]

@vivekreddynv this change cannot be cherry-picked to 202012 branch cleanly, please create a separate PR for 202012 branch.

Yeah that's probably because this change is made on the top of #5725 and that commit is reverted #5725 Revert for 202012 branch

So, to port this change to 202012, you can try reverting this commit #5725 Revert and then try cherry picking this change in 202012.

I will do it and raise a PR against 202012