Improving kexec boot time

on YouTube ↗44m

" summary

Kernel boot time drops from about 4s to ~1s in tests by enabling deferred struct page init and applying parallel SMP boot so secondary CPUs bring themselves online together; HugeTLB tail-page work (HVO) avoids initializing ~260k identical tail struct pages per 1GiB hugepage and recovers ~3s, PCI-probe filtering saves ~120ms (~8%), and disabling purgatory saves ~200ms (~20% of downtime); remaining hotspots are ~300ms of ACPI and ~100ms of necessary struct-page init, leaving VM downtime around 1.5–2s.

AI-written summary · about

" chapters (29)

1. 0:00Introduction to boot time
2. 1:04Why improve kernel boot time
3. 4:30Live update vs live migration
4. 6:15Measuring downtime and boot time
5. 7:41Initial boot time breakdown
6. 8:22Optimizing struct page initialization
7. 9:03Understanding SMP boot process
8. 10:29Parallelizing BP kick AP stage
9. 11:32Results of parallel SMP boots
10. 12:14Huge pages and boot time impact
11. 13:15Huge page VM map optimization
12. 14:20Final downtime and remaining challenges
13. 15:21Purgatory and integrity checks
14. 17:06PCI device filtering optimization
15. 17:48ACPI operations and remaining bottlenecks
16. 19:38Future work and upstream discussions
17. 21:01ACPI CPU object initialization
18. 22:04PCI state preservation across kexec
19. 24:11Kexec boot methods and purgatory
20. 25:57Struct page initialization details
21. 27:03Lazy struct page initialization
22. 28:07Asynchronous KVM and VCPU handling
23. 30:13Passing state between kernels
24. 34:10Standardized state passing framework
25. 36:41Driver serialization and PCI challenges
26. 37:22Keeping guest running during kexec
27. 39:08EPT and memory slot challenges
28. 40:10PKRAM and memory fragmentation
29. 43:30Summary and closing remarks

" related talks

• 24mSoftware Optimization for Early Linux Boot for Time Critical...Soumya Tripathy, Aashvij Shenai · Embedded Linux Conference @ EOSS NA 2024
• 38mDeferred Memblocks Init for Boot Time ReductionSudarshan Rajagopalan · Embedded Open Source Summit 2024
• 48mThe Cost of Security: Measuring and Reducing Boot-Time ImpactMichael Olbrich · Open Source Summit Europe 2025
• 15mThe Practice Method to Speed Up 10x Boot-up Time for Guest in Alibaba CloudWeinan Li · KVM Forum 2020
• 25mExploring Intel Sub-Page Protection in Context of Live MigrationManish Mishra, Shivam Kumar · KVM Forum 2022
• 20mSpeed Up Creation of a VM With Pass Through GPULiang Li · KVM Forum 2020
• 42mWhen the OS gets in the wayMark Price · Strange Loop 2015
• 40mAchieving the Ultimate Performance with KVMBoyan Krosnov · SREcon20 Americas