[PATCH 0/3] mm: page_counter: move hierarchical protection out of struct page_counter

Hierarchical memory protection (memory.min / memory.low) is built on
struct page_counter today: every counter carries the full protection
state - emin/elow, the protected-usage trackers (min_usage,
children_min_usage, low_usage, children_low_usage), the configured
min/low values and a protection_support flag - although only the memory
page counter (and dmem pools) ever participates in protection.
swap/memsw, kmem, tcpmem and hugetlb counters ship this state around
unused.

This series moves that state into a dedicated struct
page_counter_protection, instantiated only for the counters that
actually support protection, which shrinks struct page_counter by one
cache line.

Patch 1 adds struct page_counter_protection and links it to struct
page_counter through a ->prot pointer (NULL when protection is not
supported).  page_counter_init() loses its protection_support argument,
and the new page_counter_init_protection() attaches the context.
Protection stays enabled only on the cgroup v2 hierarchy, matching the
previous page_counter_init(..., memcg_on_dfl) behaviour, and the root
memcg keeps it unconditionally.

Patch 2 migrates the read/write side of protection onto the new
structure: propagate_protected_usage(), page_counter_set_min()/low()
and page_counter_calculate_protection() now operate on the protection
context, and the memcg and dmem accessors (mem_cgroup_protection,
mem_cgroup_below_min/low, the dmem below_min/low helpers and the dmem
eviction check) read emin/elow/children_*_usage from it.

Patch 3 deletes the now-unused fields from struct page_counter.  On
64-bit the structure drops from three cache lines to two, one cache
line saved per counter.  For reference, pahole shows the layout before
and after (x86_64, 64-byte cache lines):

    before:                                after:
     0  usage                               0  usage
     8  failcnt                             8  failcnt
    64  emin                               64  watermark
    72  min_usage                          72  local_watermark
    80  children_min_usage                 80  track_failcnt
    88  elow                               88  high
    96  low_usage                          96  max
   104  children_low_usage                104  parent
   112  watermark                         112  prot
   120  local_watermark
   128  protection_support                 size 128, 2 cachelines,
   129  track_failcnt                         11 members
   136  min
   144  low                                (the protection fields moved
   152  high                               into struct
   160  max                                page_counter_protection,
   168  parent                              72 bytes, allocated only
                                           where protection is used)
   size 192, 3 cachelines, 19 members

The four embedded page counters of struct mem_cgroup all shrink by 64
bytes, which translates to 128 bytes saved per cgroup once the one
embedded page_counter_protection is accounted for (2176 -> 2048 bytes
with CONFIG_MEMCG_V1=y, verified with pahole).

No functional change is intended: protection semantics and the cgroup
v1/v2 behaviour are preserved.

Signed-off-by: linuszeng <redacted>
---
linuszeng (3):
      mm: page_counter: add page_counter_protection struct and init API
      mm: page_counter: track protection state in page_counter_protection
      mm: page_counter: drop protection fields from struct page_counter

 include/linux/memcontrol.h   | 15 ++++++---
 include/linux/page_counter.h | 76 ++++++++++++++++++++++++++++++++------------
 kernel/cgroup/dmem.c         | 18 ++++++-----
 mm/hugetlb_cgroup.c          |  4 +--
 mm/memcontrol.c              | 29 ++++++++++-------
 mm/page_counter.c            | 61 ++++++++++++++++++++++-------------
 6 files changed, 134 insertions(+), 69 deletions(-)
---
base-commit: d118502628f8b673be9023db8bdf878f64a7ed45
change-id: 20260909-descriptive-name-e382a3f978dd

Best regards,
-- 
linuszeng [off-list ref]