[RFC PATCH V1 4/6] sched/numa: Increase tasks' access history

[Raghavendra K T <raghavendra.kt@amd.com>]

From: Peter Zijlstra <peterz@infradead.org>

from two to four.

This prepares for optimizations based on tasks' VMA access history.

Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Raghavendra K T <raghavendra.kt@amd.com>
---
 include/linux/mm.h       | 12 ++++++++----
 include/linux/mm_types.h |  4 +++-
 kernel/sched/fair.c      | 29 ++++++++++++++++++++++++-----
 3 files changed, 35 insertions(+), 10 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 406ab9ea818f..7794dc91c50f 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1689,10 +1689,14 @@ static inline int xchg_page_access_time(struct page *page, int time)
 static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)
 {
 	unsigned int pid_bit;
-
-	pid_bit = hash_32(current->pid, ilog2(BITS_PER_LONG));
-	if (vma->numab_state && !test_bit(pid_bit, &vma->numab_state->access_pids[1])) {
-		__set_bit(pid_bit, &vma->numab_state->access_pids[1]);
+	unsigned long *pids, pid_idx;
+
+	if (vma->numab_state) {
+		pid_bit = hash_32(current->pid, ilog2(BITS_PER_LONG));
+		pid_idx = READ_ONCE(vma->numab_state->access_pid_idx);
+		pids = vma->numab_state->access_pids + pid_idx;
+		if (!test_bit(pid_bit, pids))
+			__set_bit(pid_bit, pids);
 	}
 }
 #else /* !CONFIG_NUMA_BALANCING */
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index 647d9fc5da8d..676afa9e497c 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -475,10 +475,12 @@ struct vma_lock {
 	struct rw_semaphore lock;
 };
 
+#define NR_ACCESS_PID_HIST	4
 struct vma_numab_state {
 	unsigned long next_scan;
 	unsigned long next_pid_reset;
-	unsigned long access_pids[2];
+	unsigned long access_pids[NR_ACCESS_PID_HIST];
+	unsigned long access_pid_idx;
 	unsigned long vma_scan_select;
 };
 
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index e26e847a8e26..3ae2a1a3ef5c 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2958,12 +2958,26 @@ static bool task_disjoint_vma_select(struct vm_area_struct *vma)
 	return true;
 }
 
+static inline bool vma_test_access_pid_history(struct vm_area_struct *vma)
+{
+	unsigned int i, pid_bit;
+	unsigned long pids = 0;
+
+	pid_bit = hash_32(current->pid, ilog2(BITS_PER_LONG));
+
+	for (i = 0; i < NR_ACCESS_PID_HIST; i++)
+		pids  |= vma->numab_state->access_pids[i];
+
+	return test_bit(pid_bit, &pids);
+}
+
 static bool vma_is_accessed(struct vm_area_struct *vma)
 {
-	unsigned long pids;
+	/* Check if the current task had historically accessed VMA. */
+	if (vma_test_access_pid_history(vma))
+		return true;
 
-	pids = vma->numab_state->access_pids[0] | vma->numab_state->access_pids[1];
-	return test_bit(hash_32(current->pid, ilog2(BITS_PER_LONG)), &pids);
+	return false;
 }
 
 #define VMA_PID_RESET_PERIOD (4 * sysctl_numa_balancing_scan_delay)
@@ -2983,6 +2997,7 @@ static void task_numa_work(struct callback_head *work)
 	unsigned long nr_pte_updates = 0;
 	long pages, virtpages;
 	struct vma_iterator vmi;
+	unsigned long pid_idx;
 
 	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
 
@@ -3097,8 +3112,12 @@ static void task_numa_work(struct callback_head *work)
 				time_after(jiffies, vma->numab_state->next_pid_reset)) {
 			vma->numab_state->next_pid_reset = vma->numab_state->next_pid_reset +
 				msecs_to_jiffies(VMA_PID_RESET_PERIOD);
-			vma->numab_state->access_pids[0] = READ_ONCE(vma->numab_state->access_pids[1]);
-			vma->numab_state->access_pids[1] = 0;
+
+			pid_idx = vma->numab_state->access_pid_idx;
+			pid_idx = (pid_idx + 1) % NR_ACCESS_PID_HIST;
+
+			vma->numab_state->access_pid_idx = pid_idx;
+			vma->numab_state->access_pids[pid_idx] = 0;
 		}
 
 		/*
-- 
2.34.1