LKML News v5.12-rc3

[PATCH 00/10] [v6] Migrate Pages in lieu of discard (Dave Hansen)

https://lkml.kernel.org/r/20210304235949.7922C1C3@viggo.jf.intel.com

When the system configured with both DRAM and persistent memory, current kernel starts reclamation when DRAM fulls even though there are some space in persistent memory. And from some point, all page allocation is done on the persistent memory only, even though DRAM has some free slots now.

This patchset mitigates the problem by migrating pages that about to be reclaimed to persistent memory. The migration destination could be any slow tier memory.

[PATCH RFCv2] mm/madvise: introduce MADV_POPULATE_(READ|WRITE) to prefault/prealloc memory (David Hildenbrand)

https://lkml.kernel.org/r/20210308164520.18323-1-david@redhat.com

For user space sparse memory mappings management (e.g., hypervisors for memory ballooning or similar techniques or memory allocators), dynamic population and discard of such a sparse memory region is beneficial. This commit implements two new madvise() hints for the purpose.

[PATCH v4 0/5] Allocate memmap from hotadded memory (per device) (Oscar Salvador)

https://lkml.kernel.org/r/20210309175546.5877-1-osalvador@suse.de

This patchset reduces memory overhead due to hot-added memory, for SPARSEMEM_VMEMMAP memory model. The main three problems are, we are consuming additional memory even before hot-added memory is onlined, memmap could located on a different numa node, and memmap could populated with base pages only if the memory is fragmented. This patchset mitigates the problems by allocating the memmap from the hot-added memory itself.

[PATCH 0/5] Introduce a bulk order-0 page allocator with two in-tree users (Mel Gorman)

https://lkml.kernel.org/r/20210310104618.22750-1-mgorman@techsingularity.net

This patchset introduces a bulk order-0 page allocator and make sunrpc and network pagepool to be the users of it.

[PATCH v4 0/4] Make alloc_contig_range handle Hugetlb pages (Oscar Salvador)

https://lkml.kernel.org/r/20210310150853.13541-1-osalvador@suse.de

This commit let alloc_contig_range() which allocates physically contiguous memory, to handle HugeTLB pages for better flexibility.

[RFC -V6 0/6] NUMA balancing: optimize memory placement for memory tiering system (Huang Ying)

https://lkml.kernel.org/r/20210311081821.138467-1-ying.huang@intel.com

This is a followup of not-yet-merged patchset, ‘Migrate Pages in lieu of discard’. These are part of the pmem-based numa balancing. That is, placing hot pages in DRAM and cold pages in pmem node.

[PATCH v1 00/14] Multigenerational LRU (Yu Zhao)

https://lkml.kernel.org/r/20210313075747.3781593-1-yuzhao@google.com

This patchset makes the page reclamation logic to make finer-grained eviction target decision, by maintaining multiple LRU lists based on their age. The aging is done by finding newly referenced pages via page table scanning. After that, when eviction is needed, it selects the eviction target pages from the oldest lru list.

They used this approach on Chrome OS and achieved 96% fewer low-memory tab discards and 59% fewer OOM kills.

Linux 5.12-rc3 (Linus Torvalds)

https://lkml.kernel.org/r/CAHk-=wgAr4Z2deEQs+5L6bJb68FouwBZUSURh+m-47TBnEsGZg@mail.gmail.com

This round of the release is pretty big compared to other rc3, but it’s due to early release of rc2. There are also many rebased commits that made due to the swap file bug in rc1, which marked as recent unnecessarily. Excluding such things, this release is actually smaller than usual, Torvalds says.

Below is the diffstat of the releases in the last two years.

Note that the y-axis is in logarithm. I draw it using https://github.com/sjp38/relstat and https://github.com/sjp38/lazybox using below command:

$ relstat.py --since 2019-03-15 | ~/lazybox/gnuplot/plot.py \
    --data_fmt table --type labeled-lines --xtics_rotate -90 \
    --font "Times New Roman, 5pt" --ylog --pointsize 0.3

And, below is the diffstat of the -rc3 releases in the last two years.