Overview¶

A few notes about the KSM scanning process,to make it easier to understand the data structures below:

In order to reduce excessive scanning, KSM sorts the memory pages by theircontents into a data structure that holds pointers to the pages’ locations.

Since the contents of the pages may change at any moment, KSM cannot justinsert the pages into a normal sorted tree and expect it to find anything.Therefore KSM uses two data structures - the stable and the unstable tree.

The stable tree holds pointers to all the merged pages (ksm pages), sortedby their contents. Because each such page is write-protected, searching onthis tree is fully assured to be working (except when pages are unmapped),and therefore this tree is called the stable tree.

The stable tree node includes information required for reversemapping from a KSM page to virtual addresses that map this page.

In order to avoid large latencies of the rmap walks on KSM pages,KSM maintains two types of nodes in the stable tree:

• the regular nodes that keep the reverse mapping structures in alinked list

• the “chains” that link nodes (“dups”) that represent the samewrite protected memory content, but each “dup” corresponds to adifferent KSM page copy of that content

Internally, the regular nodes, “dups” and “chains” are representedusing the samestructksm_stable_node structure.

In addition to the stable tree, KSM uses a second data structure called theunstable tree: this tree holds pointers to pages which have been found tobe “unchanged for a period of time”. The unstable tree sorts these pagesby their contents, but since they are not write-protected, KSM cannot relyupon the unstable tree to work correctly - the unstable tree is liable tobe corrupted as its contents are modified, and so it is called unstable.

KSM solves this problem by several techniques:

1. The unstable tree is flushed every time KSM completes scanning allmemory areas, and then the tree is rebuilt again from the beginning.

2. KSM will only insert into the unstable tree, pages whose hash valuehas not changed since the previous scan of all memory areas.

3. The unstable tree is a RedBlack Tree - so its balancing is based on thecolors of the nodes and not on their contents, assuring that even whenthe tree gets “corrupted” it won’t get out of balance, so scanning timeremains the same (also, searching and inserting nodes in an rbtree usesthe same algorithm, so we have no overhead when we flush and rebuild).

4. KSM never flushes the stable tree, which means that even if it were totake 10 attempts to find a page in the unstable tree, once it is found,it is secured in the stable tree. (When we scan a new page, we firstcompare it against the stable tree, and then against the unstable tree.)

If the merge_across_nodes tunable is unset, then KSM maintains multiplestable trees and multiple unstable trees: one of each for each NUMA node.

Reverse mapping¶

KSM maintains reverse mapping information for KSM pages in the stabletree.

If a KSM page is shared between less thanmax_page_sharing VMAs,the node of the stable tree that represents such KSM page points to alist ofstructksm_rmap_item and thepage->mapping of theKSM page points to the stable tree node.

When the sharing passes this threshold, KSM adds a second dimension tothe stable tree. The tree node becomes a “chain” that links one ormore “dups”. Each “dup” keeps reverse mapping information for a KSMpage withpage->mapping pointing to that “dup”.

Every “chain” and all “dups” linked into a “chain” enforce theinvariant that they represent the same write protected memory content,even if each “dup” will be pointed by a different KSM page copy ofthat content.

This way the stable tree lookup computational complexity is unaffectedif compared to an unlimited list of reverse mappings. It is stillenforced that there cannot be KSM page content duplicates in thestable tree itself.

The deduplication limit enforced bymax_page_sharing is requiredto avoid the virtual memory rmap lists to grow too large. The rmapwalk has O(N) complexity where N is the number of rmap_items(i.e. virtual mappings) that are sharing the page, which is in turncapped bymax_page_sharing. So this effectively spreads the linearO(N) computational complexity from rmap walk context over differentKSM pages. The ksmd walk over the stable_node “chains” is also O(N),but N is the number of stable_node “dups”, not the number ofrmap_items, so it has not a significant impact on ksmd performance. Inpractice the best stable_node “dup” candidate will be kept and foundat the head of the “dups” list.

High values ofmax_page_sharing result in faster memory merging(because there will be fewer stable_node dups queued into thestable_node chain->hlist to check for pruning) and higherdeduplication factor at the expense of slower worst case for rmapwalks for any KSM page which can happen during swapping, compaction,NUMA balancing and page migration.

Thestable_node_dups/stable_node_chains ratio is also affected by themax_page_sharing tunable, and an high ratio may indicate fragmentationin the stable_node dups, which could be solved by introducingfragmentation algorithms in ksmd which would refile rmap_items fromone stable_node dup to another stable_node dup, in order to free upstable_node “dups” with few rmap_items in them, but that may increasethe ksmd CPU usage and possibly slowdown the readonly computations onthe KSM pages of the applications.

The whole list of stable_node “dups” linked in the stable_node“chains” is scanned periodically in order to prune stale stable_nodes.The frequency of such scans is defined bystable_node_chains_prune_millisecs sysfs tunable.

Reference¶

structksm_scan¶

cursor for scanning

struct ksm_scan {    struct ksm_mm_slot *mm_slot;    unsigned long address;    struct ksm_rmap_item **rmap_list;    unsigned long seqnr;};

Description

There is only the one ksm_scan instance of this cursor structure.

--Izik Eidus,Hugh Dickins, 17 Nov 2009