[PATCH v10 06/14] mm: multi-gen LRU: minimal implementation

[Yu Zhao]

On Mon, Apr 18, 2022 at 3:58 AM Barry Song <21cnbao at gmail.com> wrote:
>
> On Thu, Apr 7, 2022 at 3:16 PM Yu Zhao <yuzhao at google.com> wrote:
> >
> > To avoid confusion, the terms "promotion" and "demotion" will be
> > applied to the multi-gen LRU, as a new convention; the terms
> > "activation" and "deactivation" will be applied to the active/inactive
> > LRU, as usual.
> >
> > The aging produces young generations. Given an lruvec, it increments
> > max_seq when max_seq-min_seq+1 approaches MIN_NR_GENS. The aging
> > promotes hot pages to the youngest generation when it finds them
> > accessed through page tables; the demotion of cold pages happens
> > consequently when it increments max_seq. The aging has the complexity
> > O(nr_hot_pages), since it is only interested in hot pages. Promotion
> > in the aging path does not require any LRU list operations, only the
> > updates of the gen counter and lrugen->nr_pages[]; demotion, unless as
> > the result of the increment of max_seq, requires LRU list operations,
> > e.g., lru_deactivate_fn().
> >
> > The eviction consumes old generations. Given an lruvec, it increments
> > min_seq when the lists indexed by min_seq%MAX_NR_GENS become empty. A
> > feedback loop modeled after the PID controller monitors refaults over
> > anon and file types and decides which type to evict when both types
> > are available from the same generation.
> >
> > Each generation is divided into multiple tiers. Tiers represent
> > different ranges of numbers of accesses through file descriptors. A
> > page accessed N times through file descriptors is in tier
> > order_base_2(N). Tiers do not have dedicated lrugen->lists[], only
> > bits in folio->flags. In contrast to moving across generations, which
> > requires the LRU lock, moving across tiers only involves operations on
> > folio->flags. The feedback loop also monitors refaults over all tiers
> > and decides when to protect pages in which tiers (N>1), using the
> > first tier (N=0,1) as a baseline. The first tier contains single-use
> > unmapped clean pages, which are most likely the best choices. The
> > eviction moves a page to the next generation, i.e., min_seq+1, if the
> > feedback loop decides so. This approach has the following advantages:
> > 1. It removes the cost of activation in the buffered access path by
> >    inferring whether pages accessed multiple times through file
> >    descriptors are statistically hot and thus worth protecting in the
> >    eviction path.
> > 2. It takes pages accessed through page tables into account and avoids
> >    overprotecting pages accessed multiple times through file
> >    descriptors. (Pages accessed through page tables are in the first
> >    tier, since N=0.)
> > 3. More tiers provide better protection for pages accessed more than
> >    twice through file descriptors, when under heavy buffered I/O
> >    workloads.
> >
>
> Hi Yu,
> As I told you before,  I tried to change the current LRU (not MGLRU) by only
> promoting unmapped file pages to the head of the inactive head rather than
> the active head on its second access:
> https://lore.kernel.org/lkml/CAGsJ_4y=TkCGoWWtWSAptW4RDFUEBeYXwfwu=fUFvV4Sa4VA4A@mail.gmail.com/
> I have already seen some very good results by the decease of cpu consumption of
> kswapd and direct reclamation in the testing.

Glad to hear. I suspected you'd see some good results with that change :)

> in mglru, it seems "twice" isn't a concern at all, one unmapped file
> page accessed
> twice has no much difference with those ones which are accessed once as you
> only begin to increase refs from the third time:

refs are *additional* accesses:
PG_referenced: N=1
PG_referenced+PG_workingset: N=2
PG_referenced+PG_workingset+refs: N=3,4,5

When N=2, order_base_2(N)=1. So pages accessed twice are in the second
tier. Therefore they are "different".

More details [1]:

+/*
+ * Each generation is divided into multiple tiers. Tiers represent different
+ * ranges of numbers of accesses through file descriptors. A page accessed N
+ * times through file descriptors is in tier order_base_2(N). A page in the
+ * first tier (N=0,1) is marked by PG_referenced unless it was faulted in
+ * though page tables or read ahead. A page in any other tier (N>1) is marked
+ * by PG_referenced and PG_workingset.
+ *
+ * In contrast to moving across generations which requires the LRU lock, moving
+ * across tiers only requires operations on folio->flags and therefore has a
+ * negligible cost in the buffered access path. In the eviction path,
+ * comparisons of refaulted/(evicted+protected) from the first tier and the
+ * rest infer whether pages accessed multiple times through file descriptors
+ * are statistically hot and thus worth protecting.
+ *
+ * MAX_NR_TIERS is set to 4 so that the multi-gen LRU can support twice of the
+ * categories of the active/inactive LRU when keeping track of accesses through
+ * file descriptors. It requires MAX_NR_TIERS-2 additional bits in
folio->flags.
+ */
+#define MAX_NR_TIERS 4U

[1] https://lore.kernel.org/linux-mm/20220407031525.2368067-7-yuzhao@google.com/

> +static void folio_inc_refs(struct folio *folio)
> +{
> +       unsigned long refs;
> +       unsigned long old_flags, new_flags;
> +
> +       if (folio_test_unevictable(folio))
> +               return;
> +
> +       /* see the comment on MAX_NR_TIERS */
> +       do {
> +               new_flags = old_flags = READ_ONCE(folio->flags);
> +
> +               if (!(new_flags & BIT(PG_referenced))) {
> +                       new_flags |= BIT(PG_referenced);
> +                       continue;
> +               }
> +
> +               if (!(new_flags & BIT(PG_workingset))) {
> +                       new_flags |= BIT(PG_workingset);
> +                       continue;
> +               }
> +
> +               refs = new_flags & LRU_REFS_MASK;
> +               refs = min(refs + BIT(LRU_REFS_PGOFF), LRU_REFS_MASK);
> +
> +               new_flags &= ~LRU_REFS_MASK;
> +               new_flags |= refs;
> +       } while (new_flags != old_flags &&
> +                cmpxchg(&folio->flags, old_flags, new_flags) != old_flags);
> +}
>
> So my question is what makes you so confident that twice doesn't need
> any special treatment while the vanilla kernel is upgrading this kind of page
> to the head of the active instead? I am asking this because I am considering
> reclaiming unmapped file pages which are only accessed twice when they
> get to the tail of the inactive list.

Per above, pages accessed twice are in their own tier. Hope this clarifies it.