git.openwrt.org Git - openwrt/staging/robimarko.git/blob

   1 From 5ddf9d53d375e42af49b744bd7c2f8247c6bce15 Mon Sep 17 00:00:00 2001
   2 From: "T.J. Alumbaugh" <talumbau@google.com>
   3 Date: Wed, 18 Jan 2023 00:18:22 +0000
   4 Subject: [PATCH 13/19] UPSTREAM: mm: multi-gen LRU: section for rmap/PT walk
   5  feedback
   6
   7 Add a section for lru_gen_look_around() in the code and the design doc.
   8
   9 Link: https://lkml.kernel.org/r/20230118001827.1040870-3-talumbau@google.com
  10 Change-Id: I5097af63f61b3b69ec2abee6cdbdc33c296df213
  11 Signed-off-by: T.J. Alumbaugh <talumbau@google.com>
  12 Cc: Yu Zhao <yuzhao@google.com>
  13 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  14 (cherry picked from commit db19a43d9b3a8876552f00f656008206ef9a5efa)
  15 Bug: 274865848
  16 Signed-off-by: T.J. Mercier <tjmercier@google.com>
  17 ---
  18  Documentation/mm/multigen_lru.rst | 14 ++++++++++++++
  19  mm/vmscan.c                       |  4 ++++
  20  2 files changed, 18 insertions(+)
  21
  22 diff --git a/Documentation/mm/multigen_lru.rst b/Documentation/mm/multigen_lru.rst
  23 index 6e1483e70fdca..bd988a142bc2f 100644
  24 --- a/Documentation/mm/multigen_lru.rst
  25 +++ b/Documentation/mm/multigen_lru.rst
  26 @@ -156,6 +156,20 @@ This time-based approach has the following advantages:
  27     and memory sizes.
  28  2. It is more reliable because it is directly wired to the OOM killer.
  29
  30 +Rmap/PT walk feedback
  31 +---------------------
  32 +Searching the rmap for PTEs mapping each page on an LRU list (to test
  33 +and clear the accessed bit) can be expensive because pages from
  34 +different VMAs (PA space) are not cache friendly to the rmap (VA
  35 +space). For workloads mostly using mapped pages, searching the rmap
  36 +can incur the highest CPU cost in the reclaim path.
  37 +
  38 +``lru_gen_look_around()`` exploits spatial locality to reduce the
  39 +trips into the rmap. It scans the adjacent PTEs of a young PTE and
  40 +promotes hot pages. If the scan was done cacheline efficiently, it
  41 +adds the PMD entry pointing to the PTE table to the Bloom filter. This
  42 +forms a feedback loop between the eviction and the aging.
  43 +
  44  Summary
  45  -------
  46  The multi-gen LRU can be disassembled into the following parts:
  47 diff --git a/mm/vmscan.c b/mm/vmscan.c
  48 index ab0b8d3b9d88f..8fa82630240d6 100644
  49 --- a/mm/vmscan.c
  50 +++ b/mm/vmscan.c
  51 @@ -4553,6 +4553,10 @@ static void lru_gen_age_node(struct pglist_data *pgdat, struct scan_control *sc)
  52         }
  53  }
  54
  55 +/******************************************************************************
  56 + *                          rmap/PT walk feedback
  57 + ******************************************************************************/
  58 +
  59  /*
  60   * This function exploits spatial locality when shrink_folio_list() walks the
  61   * rmap. It scans the adjacent PTEs of a young PTE and promotes hot pages. If
  62 --
  63 2.40.1
  64