~canonical-kernel-team/+git/autotest-client-tests:jacobmartin/expect-kver-specific-mofed-mods

Last commit made on 2024-06-20
Get this branch:
git clone -b jacobmartin/expect-kver-specific-mofed-mods https://git.launchpad.net/~canonical-kernel-team/+git/autotest-client-tests
Members of Canonical Kernel Team can upload to this branch. Log in for directions.

Branch merges

Branch information

Name:
jacobmartin/expect-kver-specific-mofed-mods
Repository:
lp:~canonical-kernel-team/+git/autotest-client-tests

Recent commits

ce78d85... by Jacob Martin

UBUNTU: SAUCE: ubuntu_dgx_mofed_build: Add expected modules list for MOFED 23.10-1.1.9.0 on K6.5 jammy backports

Signed-off-by: Jacob Martin <email address hidden>

e499cd7... by Jacob Martin

UBUNTU: SAUCE: ubuntu_dgx_mofed_build: Allow expected module lists to target a specific major kernel version

The MOFED DKMS package can build additional modules when certain kernel
configs are enabled. For example, our 6.5 and 6.8 Nvidia kernels have
CONFIG_MLX5_VFIO_PCI enabled, while 5.15 does not have the option at
all. With it enabled, the MOFED DKMS packages build the
"mlx5-vfio-pci.ko" kernel module, which causes the test to fail with our
current list meant for K5.15.

Resolve this by allowing expected module lists to be specified for a
particular major kernel version, while falling back to the less specific
list if a specific one doesn't exist.

Signed-off-by: Jacob Martin <email address hidden>

a08d448... by Po-Hsu Lin

UBUNTU: SAUCE: ubuntu_performance_multipass: only calculate stddev with 2+ results

BugLink: https://bugs.launchpad.net/bugs/2067978

The formula for sample standard deviation require more than 1 result,
otherwise it will cause a ZeroDivisionError.

Calculate stddev only when we have 2+ results available to avoid this issue.

Signed-off-by: Po-Hsu Lin <email address hidden>
Acked-by: Francis Ginther <email address hidden>
Signed-off-by: Po-Hsu Lin <email address hidden>

ed3325d... by Po-Hsu Lin

UBUNTU: SAUCE: ubuntu_ltp_stable: update timeout for hugemmap06, hugemmap15

BugLink: https://bugs.launchpad.net/bugs/2068006

Issue found on ARM64 node appleton-kernel.
hugemmap06 test took 48 seconds to run with N-generic-64k 6.8.0-35.35
hugemmap15 test took 518 seconds to run with J-generic-64k 5.15.0-112.122

Bump the timeout multiplier to 5x and 25x respectively. So that it will
be 150 seconds and 750 seconds.

Signed-off-by: Po-Hsu Lin <email address hidden>
Acked-by: Francis Ginther <email address hidden>
Signed-off-by: Po-Hsu Lin <email address hidden>

6a1e76d... by Jacob Martin

UBUNTU: SAUCE: ubuntu_performance_fio: add [rd|wr]_bandwidth_gib_per_sec output value keys

DGX performance reports expect GiB/s units on the output bandwidth.
Typically fio happens to output using these units, but on some tests the
values are low enough to be output in MiB/s. Provide a new GiB/s output
value key to avoid breaking any existing tooling while providing an
option with a stable unit.

Signed-off-by: Jacob Martin <email address hidden>
Acked-by: Po-Hsu Lin <email address hidden>
Signed-off-by: Po-Hsu Lin <email address hidden>

803746c... by Po-Hsu Lin

UBUNTU: SAUCE: copy forward blacklist for Oracular

Copy the blacklist from Noble to Oracular by:
  find ./ -name blacklist.noble

Signed-off-by: Po-Hsu Lin <email address hidden>

485aafd... by Masahiro Yamada

UBUNTU: SAUCE: ubuntu_kernel_selftests: disable ptrace/get_set_sud if CONFIG_GENERIC_ENTRY=n

BugLink: https://bugs.launchpad.net/bugs/2044079

The ptrace requests:

  PTRACE_SET_SYSCALL_USER_DISPATCH_CONFIG
  PTRACE_GET_SYSCALL_USER_DISPATCH_CONFIG

... are supported only for architectures with CONFIG_GENERIC_ENTRY=y,
which are x86, s390, riscv, loongarch as of Linux 6.8.

For the other architectures, the ptrace/get_set_sud kselftest always
fails. It is tedious to give a hint to every platform.

Check /boot/config-$(uname), and if CONFIG_GENERIC_ENTRY=y is not found,
let's disable this test.

Signed-off-by: Masahiro Yamada <email address hidden>
Acked-by: Po-Hsu Lin <email address hidden>
[PHLin: amend BugLink format and add test name to the title]
Signed-off-by: Po-Hsu Lin <email address hidden>

1db95c4... by Po-Hsu Lin

UBUNTU: SAUCE: ubuntu_boot: print OS info with os-release file

KERNTT-464
Use /etc/os-release for printing out OS information. So that we don't
need to have extra dependency for using lsb_release command on core
systems.

Signed-off-by: Po-Hsu Lin <email address hidden>
Acked-by: Sean Feole <email address hidden>
Acked-by: Francis Ginther <email address hidden>
Signed-off-by: Po-Hsu Lin <email address hidden>

6e3305f... by Po-Hsu Lin

UBUNTU: SAUCE: ubuntu_boot: skip apt commands on core system

KERNTT-464
There is no apt command on Ubuntu Core system. Detect if we are running
on a core system by checking for "snapd_recovery_mode" in /proc/cmdline

Package dependency will be handled by snapcraft.

Signed-off-by: Po-Hsu Lin <email address hidden>
Acked-by: Sean Feole <email address hidden>
Acked-by: Francis Ginther <email address hidden>
Signed-off-by: Po-Hsu Lin <email address hidden>

e7feaea... by Po-Hsu Lin

UBUNTU: SAUCE: ubuntu_boot: rename control file

KERNTT-464
Rename control file from control.ubuntu to control. It's the only way
to allow autotest to recognize the test without specifying where the
control file is.

With this change you can run it with something like:
  autotest/client/autotest-local run ubuntu_boot

This is essential for snap testing as we will need to ship tests
along with autotest.

Signed-off-by: Po-Hsu Lin <email address hidden>
Acked-by: Sean Feole <email address hidden>
Acked-by: Francis Ginther <email address hidden>
Signed-off-by: Po-Hsu Lin <email address hidden>