On Tue, 10 Jun 2025 at 08:26, olivia christ olivia331904@outlook.com wrote:

Hi LKFT team,

Hi Olivia,

Thanks for reaching out.

I hope all is well. I have been running kernel tests from the latest builds (kernel 5.x) on our CI environment, and there are several failures I have repeated. It looks like these failures are related to the memory management subsystem. The specific failure I continue to see relates to kmemleak not initializing on some architectures and there are also memory leak test failures.

We’ve seen similar issues in our LKFT testing — especially with recent 5.10 and 5.15 stable branches. A few things that may be affecting your runs: * kmemleak disables itself during early boot if the early log buffer overflows. This is common on ARM64 and x86 with large reserved regions. * Some recent defconfigs have CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF=y, which means you must explicitly pass kmemleak=on on the kernel command line to enable it. * Some of the memory leak failures are real.

Has anyone seen similar behaviour in the new Linaro builds or noticed a change in reliability with tests here? If so, I would appreciate any information or thoughts you may have about possible workarounds/fixes.

Yes, we’ve seen regressions in kmemleak and LTP memory tests since early May. We've been working around it by:

Passing: kmemleak=on kmemleak_early_log_size=1024

on the kernel cmdline.

Skipping the affected LTP tests using: SKIPFILE_URL=https://raw.githubusercontent.com/linaro/qa-reports-known-issues/master/ltp-...

We’ve also proposed a patch upstream to increase the early log size by default on ARM64.

Also, are there existing configuration flags or patches that we should think about turning on and/or enabling to mitigate this issue until the underlying problem is fixed?

Here are some kernel config options worth checking: CONFIG_DEBUG_KMEMLEAK=y CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF=n CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE=1024 CONFIG_MEMCG_KMEM=y

And to better understand your environment and help narrow this down, could you share a few details?

1. Which exact 5.x kernel version are you using? (e.g., 5.15.180, 5.15.102…) 2. Which tests are failing specifically? Are they from LTP? If so, which test cases? 3. What does your kernel .config look like? If you can share it, we can try to reproduce. 4. How are you building the kernel? Which defconfig and config fragments are you using? 5. Which toolchain are you using (and version)? 6. Which architectures are showing the issue? ARM64, x86, others?

Happy to help troubleshoot further once we know more about your setup.

Cheers, Anders