Re: [PATCH v5 0/9] mm: swap: mTHP swap allocator base on swap cluster order

[Kairui Song <ryncsn@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2993 bytes --]

On Thu, Aug 1, 2024 at 5:15 PM David Hildenbrand <david@redhat.com> wrote:
>
> On 31.07.24 08:49, Chris Li wrote:
> > This is the short term solutions "swap cluster order" listed
> > in my "Swap Abstraction" discussion slice 8 in the recent
> > LSF/MM conference.
> >
>
> Running the cow.c selftest on mm/mm-unstable, I get:

Hi David, thanks very much for the test and report!

>
> # [RUN] Basic COW after fork() with mprotect() optimization ... with swapped-out, PTE-mapped THP (1024 kB)
> [   51.865309] Oops: general protection fault, probably for non-canonical address 0xdead000000000108: 0000 [#1] PREEMPT SMP NOPTI
> [   51.867738] CPU: 21 UID: 0 PID: 282 Comm: kworker/21:1 Not tainted 6.11.0-rc1+ #11
> [   51.869566] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-2.fc40 04/01/2014
> [   51.871298] Workqueue: events swap_discard_work
> [   51.872211] RIP: 0010:__free_cluster+0x27/0x90
> [   51.873101] Code: 90 90 90 0f 1f 44 00 00 8b 0d 8d 95 96 01 55 48 89 fd 53 48 89 f3 85 c9 75 3a 48 8b 43 50 48 8b 4b 48 48 8d 53 48 48 83 c5 60 <48> 89 41 08 48 89 08 48 8b 45 08 48 89 55 08 48 89 43 50 48 89 6b
> [   51.876720] RSP: 0018:ffffa3dcc0aafdc8 EFLAGS: 00010286
> [   51.877752] RAX: dead000000000122 RBX: ffff8e7ed9686e00 RCX: dead000000000100
> [   51.879186] RDX: ffff8e7ed9686e48 RSI: ffff8e7ed9686e18 RDI: ffff8e7ec37831c0
> [   51.880577] RBP: ffff8e7ec5d10860 R08: 0000000000000001 R09: 0000000000000028
> [   51.881972] R10: 0000000000000200 R11: 00000000000004cb R12: ffff8e7ed9686e00
> [   51.883393] R13: 0000000000028200 R14: 0000000000028000 R15: 0000000000000000
> [   51.884827] FS:  0000000000000000(0000) GS:ffff8e822f480000(0000) knlGS:0000000000000000
> [   51.886412] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [   51.887532] CR2: 00007f37d7e17840 CR3: 0000000335a3a001 CR4: 0000000000770ef0
> [   51.888931] PKRU: 55555554
> [   51.889471] Call Trace:
> [   51.889964]  <TASK>
> [   51.890391]  ? __die_body.cold+0x19/0x27
> [   51.891174]  ? die_addr+0x3c/0x60
> [   51.891824]  ? exc_general_protection+0x14f/0x430
> [   51.892754]  ? asm_exc_general_protection+0x26/0x30
> [   51.893717]  ? __free_cluster+0x27/0x90
> [   51.894483]  ? __free_cluster+0x7e/0x90
> [   51.895245]  swap_do_scheduled_discard+0x142/0x1b0
> [   51.896189]  swap_discard_work+0x26/0x30
> [   51.896958]  process_one_work+0x211/0x5a0
> [   51.897750]  ? srso_alias_return_thunk+0x5/0xfbef5
> [   51.898693]  worker_thread+0x1c9/0x3c0
> [   51.899438]  ? __pfx_worker_thread+0x10/0x10
> [   51.900287]  kthread+0xe3/0x110
> [   51.900913]  ? __pfx_kthread+0x10/0x10
> [   51.901656]  ret_from_fork+0x34/0x50
> [   51.902377]  ? __pfx_kthread+0x10/0x10
> [   51.903114]  ret_from_fork_asm+0x1a/0x30
> [   51.903896]  </TASK>
>
>
> Maybe related to this series?

Right, I can reproduce your problem and I believe this patch can fix
it, see the attachment.

Hi Andrew, can you pick this patch too?

[-- Attachment #2: 0001-SQUASH-Fix-discard-of-full-cluster.patch --]
[-- Type: application/octet-stream, Size: 1968 bytes --]

From 31365f8d8fa5d7a30920166b8c3a60130c24412e Mon Sep 17 00:00:00 2001
From: Kairui Song <kasong@tencent.com>
Date: Thu, 1 Aug 2024 17:29:48 +0800
Subject: [PATCH] SQUASH: Fix discard of full cluster

Signed-off-by: Kairui Song <kasong@tencent.com>
---
 include/linux/swap.h | 1 +
 mm/swapfile.c        | 8 +++++---
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/include/linux/swap.h b/include/linux/swap.h
index 145e796dab84..3543ffaf982e 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -260,6 +260,7 @@ struct swap_cluster_info {
 #define CLUSTER_FLAG_FREE 1 /* This cluster is free */
 #define CLUSTER_FLAG_NONFULL 2 /* This cluster is on nonfull list */
 #define CLUSTER_FLAG_FRAG 4 /* This cluster is on nonfull list */
+#define CLUSTER_FLAG_FULL 8 /* This cluster is on full list */
 
 /*
  * The first page in the swap file is the swap header, which is always marked
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 9872e0dbfc72..5ab06a45565a 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -450,7 +450,10 @@ static void __free_cluster(struct swap_info_struct *si, struct swap_cluster_info
 	lockdep_assert_held(&si->lock);
 	lockdep_assert_held(&ci->lock);
 
-	list_move_tail(&ci->list, &si->free_clusters);
+	if (ci->flags)
+		list_move_tail(&ci->list, &si->free_clusters);
+	else
+		list_add_tail(&ci->list, &si->free_clusters);
 	ci->flags = CLUSTER_FLAG_FREE;
 	ci->order = 0;
 }
@@ -474,7 +477,6 @@ static void swap_do_scheduled_discard(struct swap_info_struct *si)
 				SWAPFILE_CLUSTER);
 
 		spin_lock(&si->lock);
-
 		spin_lock(&ci->lock);
 		__free_cluster(si, ci);
 		memset(si->swap_map + idx * SWAPFILE_CLUSTER,
@@ -666,7 +668,7 @@ static void cluster_alloc_range(struct swap_info_struct *si, struct swap_cluster
 		if (ci->flags & CLUSTER_FLAG_FRAG)
 			si->frag_cluster_nr[ci->order]--;
 		list_move_tail(&ci->list, &si->full_clusters);
-		ci->flags = 0;
+		ci->flags = CLUSTER_FLAG_FULL;
 	}
 }
 
-- 
2.45.2