linux-kernel - [RFC PATCHv2] mm: introduce reclaim throttle in MGLRU

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <20240716094348.2451312-1-zhaoyang.huang@unisoc.com>
Date: Tue, 16 Jul 2024 17:43:48 +0800
From: "zhaoyang.huang" <zhaoyang.huang@...soc.com>
To: Andrew Morton <akpm@...ux-foundation.org>, Yu Zhao <yuzhao@...gle.com>,
        <linux-mm@...ck.org>, <linux-kernel@...r.kernel.org>,
        Zhaoyang Huang
	<huangzhaoyang@...il.com>, <steve.kang@...soc.com>
Subject: [RFC PATCHv2] mm: introduce reclaim throttle in MGLRU

From: Zhaoyang Huang <zhaoyang.huang@...soc.com>

Being like legacy LRU management, direct reclaim threads could isolate
over-sized folios and then be rescheduled, which could lead to
unwanted generation update as well as the thrashing things like before.
This commit would like to have direct_reclaim be throttled by judging
the numbers of isolated and inactive folios.

This patch is verified by launching 8 costmem(malloc and access 1GB VM
in an 5.5GB v6.6 Android system) concurrently and got no system hang
any more(adb shell recovered in 10s which hanged 100% in mainline).

test script under Android14

Signed-off-by: Zhaoyang Huang <zhaoyang.huang@...soc.com>
---
v2: fix a possible unpaired spin_lock/unlock and commit message
---
---
 mm/vmscan.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 2e34de9cd0d4..13e5ed9060ad 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -4481,6 +4481,7 @@ static int isolate_folios(struct lruvec *lruvec, struct scan_control *sc, int sw
 	int scanned;
 	int tier = -1;
 	DEFINE_MIN_SEQ(lruvec);
+	bool stalled = false;
 
 	/*
 	 * Try to make the obvious choice first, and if anon and file are both
@@ -4503,6 +4504,16 @@ static int isolate_folios(struct lruvec *lruvec, struct scan_control *sc, int sw
 	else
 		type = get_type_to_scan(lruvec, swappiness, &tier);
 
+	spin_unlock_irq(&lruvec->lru_lock);
+	while (unlikely(too_many_isolated(lruvec_pgdat(lruvec), type, sc))) {
+		if (stalled) {
+			spin_lock_irq(&lruvec->lru_lock);
+			return 0;
+		}
+		reclaim_throttle(lruvec_pgdat(lruvec), VMSCAN_THROTTLE_ISOLATED);
+	}
+	spin_lock_irq(&lruvec->lru_lock);
+
 	for (i = !swappiness; i < ANON_AND_FILE; i++) {
 		if (tier < 0)
 			tier = get_tier_idx(lruvec, type);
@@ -4550,8 +4561,10 @@ static int evict_folios(struct lruvec *lruvec, struct scan_control *sc, int swap
 	if (list_empty(&list))
 		return scanned;
 retry:
+	__mod_node_page_state(lruvec_pgdat(lruvec), NR_ISOLATED_ANON + type, scanned);
 	reclaimed = shrink_folio_list(&list, pgdat, sc, &stat, false);
 	sc->nr_reclaimed += reclaimed;
+	__mod_node_page_state(lruvec_pgdat(lruvec), NR_ISOLATED_ANON + type, -scanned);
 	trace_mm_vmscan_lru_shrink_inactive(pgdat->node_id,
 			scanned, reclaimed, &stat, sc->priority,
 			type ? LRU_INACTIVE_FILE : LRU_INACTIVE_ANON);
-- 
2.25.1