[5.10][Feat][Host]Improve concurrent allocation performance for CSV3 virtual machine · Pull Request !5709 · anolis/cloud-kernel

bugzilla: https://bugzilla.openanolis.cn/show_bug.cgi?id=24405

1, Revert commit bf06164cafb1 ("anolis: mm: hugetlb: wait hugetlb pages to be released to buddy allocator") and commit 1b4d4cce1b1a ("anolis: mm: release hugetlb pages to buddy allocator after migrate them").

Using upstream commit 04f13d241b8b ("mm: replace free hugepage folios after migration") and commit 67bab13307c8 ("mm/hugetlb: wait for hugetlb folios to be freed") can also fix the bug in hugetlb page migration and improve migration performance.

2, Currently, when starting a csv3 virtual machine, the function csv_alloc_from_contiguous() is called to allocate memory, starting from csv_cma[0], then csv_cma[1], and so on in sequence. When multiple csv3 virtual machines are started concurrently, this easily causes contention on the same csv_cma, which affects performance.

It is necessary to adjust the logic so that each time the csv_alloc_from_contiguous() function is called for memory allocation, it allocates from the csv_cma that follows the one used in the previous allocation. This adjustment can significantly improve performance.

Patches:
1) mm/hugetlb: fix kernel NULL pointer dereference when replacing free hu…
2) mm/hugetlb: wait for hugetlb folios to be freed
3) mm: replace free hugepage folios after migration
4) anolis: revert: mm: release hugetlb pages to buddy allocator after migrate them
5) anolis: revert: mm: hugetlb: wait hugetlb pages to be released to budd…
6) anolis: x86/csv: Allocate memory cyclically from different csv_cma

类型	指派人员	状态
审查	Guanjun xunlei 半叶枫尘 zhongjiang Artie Ding winterddd xueshuai vxuyu baolin.wang 冯光辉 Guixin Liu health_jay kun-llfl MinLi Xuchun Shang	已完成 (0/0人)
测试	小龙已测试通过	已完成 (1/1人)

GVPanolis/cloud-kernel

搜索帮助

GVP anolis/cloud-kernel