Linux 6.19's Hung Task & System Lockup Detectors Can Provide Greater Insight
Beginning with the Linux 6.19 kernel, the hung task detector and system lock-up detector are now optionally able to provide greater insight into the issues by dumping additional system information. The new lockup_sys_info and hung_task_sys_info sysctl knobs were merged over as part of the pull requests managed by Andrew Morton.
Andrew Morton first sent in the memory management "MM" updates for Linux 6.19. Overall a number of low-level kernel code clean-ups and various minor optimizations:
- A rework to the vmalloc() code to support non-blocking allocations.
- Minor clean-ups to the NUMA node handling code. There are also small optimizations to the NUMA allocation hinting code.
- DAMOS auto-tuning has been enhanced with DAMOS auto-tuned for per-memcg per-node memory usage.
- Optimized logic for handling dirty file folios during page reclaim to reduce some unnecessary work on page reclamation.
- Optimizations to the folio splitting code.
- Support for Transparent Huge Page "THP" migration in zone device-private memory.
More details on the many low-level MM changes for Linux 6.19 via this pull request.
Following that was the "non-MM" pull from Andrew Morton with a few other interesting changes to note, including the detector dumping enhancements:
- Making BPF symbol names, sizes, and line numbers available to the GDB debugger via the BPF debug info.
- Linux's hung task detector and lockup cases can now dump additional system information on-demand. With new sysctl configuration options used to cause extra information to be dumped when the hung-task or lock-up detectors are triggered. For the hung task detector is the "hung_task_sys_info" sysctl knob that accepts a string like "tasks,mem,timers,locks,ftrace" for specifying which information should be dumped when a hung task is detected. Similarly is the "lockup_sys_info" sysctl for specifying the system information to be dumped when detecting a system lockup. Currently-supported options that can be passed to either hung_task_sys_info and/or lockup_sys_info in a comma-separated string include:
- As covered separately, the Live Update Orchestrator was merged as another big kernel contribution from Google.
More details via the non-MM pull request. Both of these pull requests have been merged.
Andrew Morton first sent in the memory management "MM" updates for Linux 6.19. Overall a number of low-level kernel code clean-ups and various minor optimizations:
- A rework to the vmalloc() code to support non-blocking allocations.
- Minor clean-ups to the NUMA node handling code. There are also small optimizations to the NUMA allocation hinting code.
- DAMOS auto-tuning has been enhanced with DAMOS auto-tuned for per-memcg per-node memory usage.
- Optimized logic for handling dirty file folios during page reclaim to reduce some unnecessary work on page reclamation.
- Optimizations to the folio splitting code.
- Support for Transparent Huge Page "THP" migration in zone device-private memory.
More details on the many low-level MM changes for Linux 6.19 via this pull request.
Following that was the "non-MM" pull from Andrew Morton with a few other interesting changes to note, including the detector dumping enhancements:
- Making BPF symbol names, sizes, and line numbers available to the GDB debugger via the BPF debug info.
- Linux's hung task detector and lockup cases can now dump additional system information on-demand. With new sysctl configuration options used to cause extra information to be dumped when the hung-task or lock-up detectors are triggered. For the hung task detector is the "hung_task_sys_info" sysctl knob that accepts a string like "tasks,mem,timers,locks,ftrace" for specifying which information should be dumped when a hung task is detected. Similarly is the "lockup_sys_info" sysctl for specifying the system information to be dumped when detecting a system lockup. Currently-supported options that can be passed to either hung_task_sys_info and/or lockup_sys_info in a comma-separated string include:
- As covered separately, the Live Update Orchestrator was merged as another big kernel contribution from Google.
More details via the non-MM pull request. Both of these pull requests have been merged.