This is an algorithmic policy. A zs_pool maintains zs_pages of different size_class. However, some size_classes share exactly same characteristics, namely pages_per_zspage and objs_per_zspage. Recall the other annotation of mine, it searched free zspages by size_class: zspage = find_get_zspage(class);. Thus, grouping different classes improves memory utilization.

This is an algorithmic policy. Instead of immediately allocating new zspages for each memory request, the algorithm first attempts to find and reuse existing partially filled zspages from a given size class by invoking the find_get_zspage(class) function. It also updates the corresponding fullness groups.

This is a configuration policy that sets the timeout between retries if vmap_pages_range() fails. This could be tunable variable.

This is an algorithmic policy. This is an optimization that prevents duplicate marks of accessibility. Only pages allocated without VM_ALLOC (e.g. ioremap) was not set accessible (unpoison), thus requiring explicit setting here.

This is an configuration policy that determines 100 pages are the upper limit for the bulk-allocator. However, the implementation of alloc_pages_bulk_array_mempolicy does not explicitly limit in the implementation. So I believe it is an algorithmic policy related to some sort of optimization.

This is an algorithmic policy determines that only use the bulk allocator for order-0 pages (non-super pages). Maybe the bulk allocator could be applied to super pages to speed up allocation. Currently I haven't seen the reason why it cannot be applied.

This is an algorithmic policy that determines whether to use a more efficient vl_alloc which does not involve complex virtual-to-physical mappings. Unlike the latter alloc_vmap_area, vb_alloc does not need to traverse the rb-tree of free vmap areas. It simply find a larger enough block from vmap_block_queue.

The threshold VMAP_PURGE_THRESHOLD is a configuration policy that could be tuned by machine learning. Setting this threshold lower reduces purging activities while setting it higher reduces framentation.

This is an algorithmic policy that prevents purging blocks with considerable amount of "usable memory" unless requested with force_purge.

The assignment of resched_threshold and lines 1776-1777 are configuration policies to determine fewer than which number of lazily-freed pages it should yield CPU temporarily to higher-priority tasks.

This heuristic scales lazy_max_pages logarithmically, which is a configuration policy. Alternatively, machine learning could determine the optimal scaling function—whether linear, logarithmic, square-root, or another approach.

This is a configuration policy that decides to always returns multiples of 32 MB worth of pages. This could be a configurable variable rather than a fixed magic number.