[RFC v3] mm: add page preemption

[Hillf Danton <hdanton@sina.com>]

The cpu preemption feature makes a task able to preempt other tasks
of lower priorities for cpu. It has been around for a while.

This work introduces task prio into page reclaiming in order to add
the page preemption feature that makes a task able to preempt other
tasks of lower priorities for page.

No page will be reclaimed on behalf of tasks of lower priorities
under pp, a two-edge feature that functions only under memory
pressure, laying a barrier to pages flowing to lower prio, and the
nice syscall is what users need to fiddle with it for instance if
they have a bunch of workloads to run in datacenter, and some
difficulty predicting the runtime working set size for every
individual workload which is sensitive to jitters in lru pages.

Currently pages are reclaimed without prio taken into account; pages
can be reclaimed from tasks of lower priorities on behalf of
higher-prio tasks and vice versa.

s/and vice versa/only/ is what we need to make pp by definition, but
it could not make a sense without prio introduced; otherwise we can
simply skip deactivating the lru pages based on prio comprison, and
work is done.

The introduction consists of two parts. On the page side, we have to
store the page owner task's prio in page, which needs an extra room the
size of the int type in the page struct. That room sounds impossible
without inflating the page struct size, and is walked around by making
pp depend on CONFIG_64BIT.

On the reclaimer side, kswapd's prio is set with the prio of its waker,
and updated in the same manner as kswapd_order.

Another change is, because pp is a two-edge option, added in oom to
avoid killing high-prio task in favor of low-prio task.

V3 is based on next-20191031.

Changes since v2
- page->prio depends on CONFIG_64BIT
- fix high-prio task killed in oom

Changes since v1
- page->prio depends on !LAST_CPUPID_NOT_IN_PAGE_FLAGS

Changes since v0
- s/page->nice/page->prio/
- drop the role of kswapd's reclaiming prioirty in prio comparison
- add pgdat->kswapd_prio

Signed-off-by: Hillf Danton <hdanton@sina.com>
---

--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -197,8 +197,18 @@ struct page {
 	/* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
 	atomic_t _refcount;
 
+	/* 32/56 bytes above */
+
 #ifdef CONFIG_MEMCG
 	struct mem_cgroup *mem_cgroup;
+#else
+#ifdef CONFIG_64BIT
+	union {
+		int prio;
+		void *__pad;
+	};
+#define CONFIG_PAGE_PREEMPTION PP
+#endif
 #endif
 
 	/*
--- /dev/null
+++ b/include/linux/page_prio.h
@@ -0,0 +1,55 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _LINUX_PAGE_PRIO_H
+#define _LINUX_PAGE_PRIO_H
+
+#include <linux/sched/prio.h>
+#include <linux/mm_types.h>
+
+#ifdef CONFIG_PAGE_PREEMPTION
+static inline bool page_prio_valid(struct page *p)
+{
+	return p->prio > MAX_PRIO;
+}
+
+static inline void set_page_prio(struct page *p, int task_prio)
+{
+	if (!page_prio_valid(p))
+		p->prio = task_prio + MAX_PRIO + 1;
+}
+
+static inline void copy_page_prio(struct page *to, struct page *from)
+{
+	to->prio = from->prio;
+}
+
+static inline int page_prio(struct page *p)
+{
+	return p->prio - MAX_PRIO - 1;
+}
+
+static inline bool page_prio_higher(struct page *p, int prio)
+{
+	return page_prio(p) < prio;
+}
+#else
+static inline bool page_prio_valid(struct page *p)
+{
+	return true;
+}
+static inline void set_page_prio(struct page *p, int task_prio)
+{
+}
+static inline void copy_page_prio(struct page *to, struct page *from)
+{
+}
+static inline int page_prio(struct page *p)
+{
+	return MAX_PRIO + 1;
+}
+static inline bool page_prio_higher(struct page *p, int prio)
+{
+	return false;
+}
+#endif /* CONFIG_PAGE_PREEMPTION */
+
+#endif /* _LINUX_PAGE_PRIO_H */
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -24,6 +24,7 @@
 #include <linux/page_ext.h>
 #include <linux/err.h>
 #include <linux/page_ref.h>
+#include <linux/page_prio.h>
 #include <linux/memremap.h>
 #include <linux/overflow.h>
 #include <linux/sizes.h>
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -671,6 +671,7 @@ static void __collapse_huge_page_copy(pt
 			}
 		} else {
 			src_page = pte_page(pteval);
+			copy_page_prio(page, src_page);
 			copy_user_highpage(page, src_page, address, vma);
 			VM_BUG_ON_PAGE(page_mapcount(src_page) != 1, src_page);
 			release_pte_page(src_page);
@@ -1735,6 +1736,7 @@ xa_unlocked:
 				clear_highpage(new_page + (index % HPAGE_PMD_NR));
 				index++;
 			}
+			copy_page_prio(new_page, page);
 			copy_highpage(new_page + (page->index % HPAGE_PMD_NR),
 					page);
 			list_del(&page->lru);
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -647,6 +647,7 @@ void migrate_page_states(struct page *ne
 		end_page_writeback(newpage);
 
 	copy_page_owner(page, newpage);
+	copy_page_prio(newpage, page);
 
 	mem_cgroup_migrate(page, newpage);
 }
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -164,6 +164,11 @@ static bool oom_unkillable_task(struct t
 		return true;
 	if (p->flags & PF_KTHREAD)
 		return true;
+
+#ifdef CONFIG_PAGE_PREEMPTION
+	if (p->prio < current->prio)
+		return true;
+#endif
 	return false;
 }
 
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1575,6 +1575,7 @@ static int shmem_replace_page(struct pag
 
 	get_page(newpage);
 	copy_highpage(newpage, oldpage);
+	copy_page_prio(newpage, oldpage);
 	flush_dcache_page(newpage);
 
 	__SetPageLocked(newpage);
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -407,6 +407,7 @@ static void __lru_cache_add(struct page
 	struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
 
 	get_page(page);
+	set_page_prio(page, current->prio);
 	if (!pagevec_add(pvec, page) || PageCompound(page))
 		__pagevec_lru_add(pvec);
 	put_cpu_var(lru_add_pvec);
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -738,6 +738,7 @@ typedef struct pglist_data {
 	int kswapd_order;
 	enum zone_type kswapd_classzone_idx;
 
+	int kswapd_prio;
 	int kswapd_failures;		/* Number of 'reclaimed == 0' runs */
 
 #ifdef CONFIG_COMPACTION
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -110,6 +110,9 @@ struct scan_control {
 	/* The highest zone to isolate pages for reclaim from */
 	s8 reclaim_idx;
 
+	s8 __pad;
+	int reclaimer_prio;
+
 	/* This context's GFP mask */
 	gfp_t gfp_mask;
 
@@ -1707,11 +1710,17 @@ static unsigned long isolate_lru_pages(u
 		total_scan += nr_pages;
 
 		if (page_zonenum(page) > sc->reclaim_idx) {
+next_page:
 			list_move(&page->lru, &pages_skipped);
 			nr_skipped[page_zonenum(page)] += nr_pages;
 			continue;
 		}
 
+#ifdef CONFIG_PAGE_PREEMPTION
+		if (is_active_lru(lru) && global_reclaim(sc) &&
+		    page_prio_higher(page, sc->reclaimer_prio))
+			goto next_page;
+#endif
 		/*
 		 * Do not count skipped pages because that makes the function
 		 * return with no isolated pages if the LRU mostly contains
@@ -3257,6 +3266,7 @@ unsigned long try_to_free_pages(struct z
 	unsigned long nr_reclaimed;
 	struct scan_control sc = {
 		.nr_to_reclaim = SWAP_CLUSTER_MAX,
+		.reclaimer_prio = current->prio,
 		.gfp_mask = current_gfp_context(gfp_mask),
 		.reclaim_idx = gfp_zone(gfp_mask),
 		.order = order,
@@ -3583,6 +3593,7 @@ static int balance_pgdat(pg_data_t *pgda
 	bool boosted;
 	struct zone *zone;
 	struct scan_control sc = {
+		.reclaimer_prio = pgdat->kswapd_prio,
 		.gfp_mask = GFP_KERNEL,
 		.order = order,
 		.may_unmap = 1,
@@ -3736,6 +3747,8 @@ restart:
 		if (nr_boost_reclaim && !nr_reclaimed)
 			break;
 
+		sc.reclaimer_prio = pgdat->kswapd_prio;
+
 		if (raise_priority || !nr_reclaimed)
 			sc.priority--;
 	} while (sc.priority >= 1);
@@ -3828,6 +3841,7 @@ static void kswapd_try_to_sleep(pg_data_
 		 */
 		wakeup_kcompactd(pgdat, alloc_order, classzone_idx);
 
+		pgdat->kswapd_prio = MAX_PRIO + 1;
 		remaining = schedule_timeout(HZ/10);
 
 		/*
@@ -3862,8 +3876,10 @@ static void kswapd_try_to_sleep(pg_data_
 		 */
 		set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold);
 
-		if (!kthread_should_stop())
+		if (!kthread_should_stop()) {
+			pgdat->kswapd_prio = MAX_PRIO + 1;
 			schedule();
+		}
 
 		set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold);
 	} else {
@@ -3914,6 +3930,7 @@ static int kswapd(void *p)
 	tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
 	set_freezable();
 
+	pgdat->kswapd_prio = MAX_PRIO + 1;
 	pgdat->kswapd_order = 0;
 	pgdat->kswapd_classzone_idx = MAX_NR_ZONES;
 	for ( ; ; ) {
@@ -3982,6 +3999,19 @@ void wakeup_kswapd(struct zone *zone, gf
 		return;
 	pgdat = zone->zone_pgdat;
 
+#ifdef CONFIG_PAGE_PREEMPTION
+	do {
+		int prio = current->prio;
+
+		if (pgdat->kswapd_prio < prio) {
+			smp_rmb();
+			return;
+		}
+		pgdat->kswapd_prio = prio;
+		smp_wmb();
+	} while (0);
+#endif
+
 	if (pgdat->kswapd_classzone_idx == MAX_NR_ZONES)
 		pgdat->kswapd_classzone_idx = classzone_idx;
 	else
--