Slightly cleaner kernel for generic VMs #5
Conversation
|
Can you include those small corrections, then try build kernel like and update modules files based build log? |
|
I have yet had no time to play with the build system. |
|
Just build with make command like it is decribed on readme. |
|
The build system seems to have an inconsistent modules list. I've purged it already 7 times... |
|
It might be that those files need to be sorted. Successfully passed build after including https://github.com/olljanat/os-kernel/commit/5f07c70e3bd3a7d8bdb7a5368bc04e5cb5277c20 and using build command: But on build result those files looks huge compared to https://github.com/burmilla/os-kernel/releases/tag/v5.10.1-burmilla |
|
We have more network cards supported as you find a lot of weird stuff in home labs. But if you want me to, I can strip it down more by throwing out some more obsolete cards (I still have drivers for a 3C509 in there) |
I mean that after your changes kernel package is ten times bigger than before these changes. Comparison of compressed files: Comparison of non-compressed files: |
|
I suspect my mistake: I'm used to embed firmware. My bad. |
| @@ -423,33 +419,34 @@ CONFIG_X86_DIRECT_GBPAGES=y | |||
| CONFIG_NUMA=y | |||
| CONFIG_AMD_NUMA=y | |||
| CONFIG_X86_64_ACPI_NUMA=y | |||
| # CONFIG_NUMA_EMU is not set | |||
| CONFIG_NUMA_EMU=y | |||
There was a problem hiding this comment.
Thinking about based on https://cateee.net/lkddb/web-lkddb/NUMA_EMU.html that does this actually make sense?
| CONFIG_X86_RESERVE_LOW=64 | ||
| CONFIG_MTRR=y | ||
| CONFIG_MTRR_SANITIZER=y | ||
| CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT=1 | ||
| CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT=0 |
There was a problem hiding this comment.
When reading this I'm not sure if we really want disable this one https://github.com/tomreyn/linux_mtrr_size_fix
| @@ -504,7 +502,7 @@ CONFIG_HIBERNATION_SNAPSHOT_DEV=y | |||
| CONFIG_PM_STD_PARTITION="" | |||
| CONFIG_PM_SLEEP=y | |||
| CONFIG_PM_SLEEP_SMP=y | |||
| # CONFIG_PM_AUTOSLEEP is not set | |||
| CONFIG_PM_AUTOSLEEP=y | |||
There was a problem hiding this comment.
I'm quite sure that we don't need support sleep for BurmillaOS so probably it makes sense to leave this untouched
and also set CONFIG_PM_SLEEP and CONFIG_PM_SLEEP_SMP to "n"
https://github.com/torvalds/linux/blob/master/kernel/power/Kconfig#L117-L145
| @@ -982,41 +1002,36 @@ CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y | |||
| # CONFIG_TRANSPARENT_HUGEPAGE_MADVISE is not set | |||
| CONFIG_ARCH_WANTS_THP_SWAP=y | |||
| CONFIG_THP_SWAP=y | |||
| CONFIG_CLEANCACHE=y | |||
| # CONFIG_CLEANCACHE is not set | |||
There was a problem hiding this comment.
Not sure if we really want disable this one https://github.com/torvalds/linux/blob/master/mm/Kconfig#L448-L468
|
It's what Debian does and I assume those guys aren't stupid. |
Yea but Debian is generic purpose OS for everything from low power laptops to high power servers which why they are forced to do compromises. BurmillaOS instead of is purpose build server OS which kernel parameters tuning Rancher Labs guys have used a lot of time which why IMO we should not change any setting unless we know that those make sense. However I cherry-picked some changes from this PR to #6 now and will start by testing those. |
|
Keep in mind what the average user at home calls a server - this can be anything from an ancient x-series up to some Intel NUC - that's why I picked some of those options. |
|
Looks that we are not able to agree best possible kernel parameters here and most probably it does not even matter as it looks to be possible share kernel with k3OS (look: burmilla/os#5 (comment) ) so future development should be done on https://github.com/rancher/k3os-kernel |
This is a 'port' of my debian-vm config which should work on HyperV, Proxmox, Virtualbox - I left a bunch of network card drivers in there as people sometimes pass through those for firewalling stuff or increasing traffic. Same for VGA cards due to plex encoding