Unable to create database on latest Linux kernel

[ftyghome]

Description of the issue

I'm attempting to analyze the Linux kernel using CodeQL, but I'm encountering errors during database creation.

Steps to reproduce:

1. Clone the latest Linux kernel repository
2. Run: codeql database create db --language c
3. CodeQL correctly detects it as a Linux kernel and starts the autobuild process
4. The build completes defconfig and compiles the kernel successfully

Problem:

The trace output contains numerous "identifier 'asm' is undefined" errors, such as:

[E 15:16:45 4019640] Warning[extractor-c++]: In construct_text_message: "./include/linux/sched/signal.h", line 278: error: identifier "asm" is undefined
        WARN_ON(sig->flags & SIGNAL_GROUP_EXIT);
        ^

"./include/linux/mm_types.h", line 1362: error: identifier "asm" is undefined
        set_mask_bits(bitmap, mask, bits);
        ^

"./arch/x86/include/asm/pgtable.h", line 741: error: identifier "asm" is undefined
        WARN_ON_ONCE((pgprot_val(pgprot) & (_PAGE_DIRTY | _PAGE_RW)) ==
        ^

Additionally, the resulting database is highly incomplete and contains very little information, making it unusable for analysis.

Environment tested:

• Ubuntu 24.04 and Arch Linux
• Latest CodeQL and version from June 2024

I would greatly appreciate any guidance on what might be causing this issue. Thank you very much for your time!

[jketema]

Hi @ftyghome,

Before diving into this, did you see https://github.com/github/codeql/issues/16908#issuecomment-2213507452? Also could you tell us exactly which CodeQL versions you are using?

[ftyghome]

Hi @ftyghome,

Before diving into this, did you see #16908 (comment)? Also could you tell us exactly which CodeQL versions you are using?

Hi @jketema,

Yes, I did see that comment. Although the issues may not be directly related, I tried following the approach mentioned in that issue and modified the kernel accordingly, but the problem persists.

I'm using CodeQL version 2.23.7 (release version).

Thank you!

[jketema]

It looks like the Linux kernel is now compiled with the -fms-extensions option https://github.com/torvalds/linux/commit/c4781dc3d1cf0e017e1f290607ddc56cfe187afc. We support this option. However, it looks like there's some issue in combination with the inline assembly that the Linux kernel uses. I'll report this problem to our frontend supplier, but I cannot give you a timeline for a fix.

[ftyghome]

Hi @jketema,

Thanks for tracking this down! Looking forward to hearing about a fix.

[jketema]

We've received a patch for the inline assembly issue. However, this uncovered a number of additional issues affecting database creation for Linux. Those have also been reported to our frontend provider.

[ftyghome]

Thanks for the update!