Re: [PATCH] mm: fix trying to reclaim unevicable LRU page

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, Jun 10, 2019 at 12:39:46PM +0200, Michal Hocko wrote:
> On Mon 10-06-19 18:42:22, Minchan Kim wrote:
> > On Tue, Jun 04, 2019 at 02:28:06PM +0200, Michal Hocko wrote:
> > > On Thu 30-05-19 11:42:29, Minchan Kim wrote:
> > > > On Tue, May 28, 2019 at 05:14:07PM +0200, Michal Hocko wrote:
> > > > > [Cc Pankaj Suryawanshi who has reported a similar problem
> > > > > http://lkml.kernel.org/r/SG2PR02MB309806967AE91179CAFEC34BE84B0@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx]
> > > > > 
> > > > > On Fri 24-05-19 16:11:14, Minchan Kim wrote:
> > > > > > There was below bugreport from Wu Fangsuo.
> > > > > > 
> > > > > > 7200 [  680.491097] c4 7125 (syz-executor) page:ffffffbf02f33b40 count:86 mapcount:84 mapping:ffffffc08fa7a810 index:0x24
> > > > > > 7201 [  680.531186] c4 7125 (syz-executor) flags: 0x19040c(referenced|uptodate|arch_1|mappedtodisk|unevictable|mlocked)
> > > > > > 7202 [  680.544987] c0 7125 (syz-executor) raw: 000000000019040c ffffffc08fa7a810 0000000000000024 0000005600000053
> > > > > > 7203 [  680.556162] c0 7125 (syz-executor) raw: ffffffc009b05b20 ffffffc009b05b20 0000000000000000 ffffffc09bf3ee80
> > > > > > 7204 [  680.566860] c0 7125 (syz-executor) page dumped because: VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page))
> > > > > > 7205 [  680.578038] c0 7125 (syz-executor) page->mem_cgroup:ffffffc09bf3ee80
> > > > > > 7206 [  680.585467] c0 7125 (syz-executor) ------------[ cut here ]------------
> > > > > > 7207 [  680.592466] c0 7125 (syz-executor) kernel BUG at /home/build/farmland/adroid9.0/kernel/linux/mm/vmscan.c:1350!
> > > > > > 7223 [  680.603663] c0 7125 (syz-executor) Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
> > > > > > 7224 [  680.611436] c0 7125 (syz-executor) Modules linked in:
> > > > > > 7225 [  680.616769] c0 7125 (syz-executor) CPU: 0 PID: 7125 Comm: syz-executor Tainted: G S              4.14.81 #3
> > > > > > 7226 [  680.626826] c0 7125 (syz-executor) Hardware name: ASR AQUILAC EVB (DT)
> > > > > > 7227 [  680.633623] c0 7125 (syz-executor) task: ffffffc00a54cd00 task.stack: ffffffc009b00000
> > > > > > 7228 [  680.641917] c0 7125 (syz-executor) PC is at shrink_page_list+0x1998/0x3240
> > > > > > 7229 [  680.649144] c0 7125 (syz-executor) LR is at shrink_page_list+0x1998/0x3240
> > > > > > 7230 [  680.656303] c0 7125 (syz-executor) pc : [<ffffff90083a2158>] lr : [<ffffff90083a2158>] pstate: 60400045
> > > > > > 7231 [  680.666086] c0 7125 (syz-executor) sp : ffffffc009b05940
> > > > > > ..
> > > > > > 7342 [  681.671308] c0 7125 (syz-executor) [<ffffff90083a2158>] shrink_page_list+0x1998/0x3240
> > > > > > 7343 [  681.679567] c0 7125 (syz-executor) [<ffffff90083a3dc0>] reclaim_clean_pages_from_list+0x3c0/0x4f0
> > > > > > 7344 [  681.688793] c0 7125 (syz-executor) [<ffffff900837ed64>] alloc_contig_range+0x3bc/0x650
> > > > > > 7347 [  681.717421] c0 7125 (syz-executor) [<ffffff90084925cc>] cma_alloc+0x214/0x668
> > > > > > 7348 [  681.724892] c0 7125 (syz-executor) [<ffffff90091e4d78>] ion_cma_allocate+0x98/0x1d8
> > > > > > 7349 [  681.732872] c0 7125 (syz-executor) [<ffffff90091e0b20>] ion_alloc+0x200/0x7e0
> > > > > > 7350 [  681.740302] c0 7125 (syz-executor) [<ffffff90091e154c>] ion_ioctl+0x18c/0x378
> > > > > > 7351 [  681.747738] c0 7125 (syz-executor) [<ffffff90084c6824>] do_vfs_ioctl+0x17c/0x1780
> > > > > > 7352 [  681.755514] c0 7125 (syz-executor) [<ffffff90084c7ed4>] SyS_ioctl+0xac/0xc0
> > > > > > 
> > > > > > Wu found it's due to [1]. Before that, unevictable page goes to cull_mlocked
> > > > > > routine so that it couldn't reach the VM_BUG_ON_PAGE line.
> > > > > > 
> > > > > > To fix the issue, this patch filter out unevictable LRU pages
> > > > > > from the reclaim_clean_pages_from_list in CMA.
> > > > > 
> > > > > The changelog is rather modest on details and I have to confess I have
> > > > > little bit hard time to understand it. E.g. why do not we need to handle
> > > > > the regular reclaim path?
> > > > 
> > > > No need to pass unevictable pages into regular reclaim patch if we are
> > > > able to know in advance.
> > > 
> > > I am sorry to be dense here. So what is the difference in the CMA path?
> > > Am I right that the pfn walk (CMA) rather than LRU isolation (reclaim)
> > > is the key differentiator?
> > 
> > Yes.
> > We could isolate unevictable LRU pages from the pfn waker to migrate and
> > could discard clean file-backed pages to reduce migration latency in CMA
> > path.
> 
> Please be explicit about that in the changelog. The fact that this is
> not possible from the regular reclaim path is really important and not
> obvious from the first glance.
> 
> Thanks!

Here it goes:

Andrew, could you replace the patch with the below if Michal doesn't have
any suggestion?

Thanks.

>From 5cb8958ea240e2580bd2b331448009e8e1854240 Mon Sep 17 00:00:00 2001
From: Minchan Kim <minchan@xxxxxxxxxx>
Date: Fri, 24 May 2019 15:54:10 +0900
Subject: [PATCH] mm: fix trying to reclaim unevicable LRU page

There was below bugreport from Wu Fangsuo.

In CMA allocation path, isolate_migratepages_range could isolate unevictable
LRU pages and reclaim_clean_page_from_list can try to reclaim them if
they are clean file-backed pages.

7200 [  680.491097] c4 7125 (syz-executor) page:ffffffbf02f33b40 count:86 mapcount:84 mapping:ffffffc08fa7a810 index:0x24
7201 [  680.531186] c4 7125 (syz-executor) flags: 0x19040c(referenced|uptodate|arch_1|mappedtodisk|unevictable|mlocked)
7202 [  680.544987] c0 7125 (syz-executor) raw: 000000000019040c ffffffc08fa7a810 0000000000000024 0000005600000053
7203 [  680.556162] c0 7125 (syz-executor) raw: ffffffc009b05b20 ffffffc009b05b20 0000000000000000 ffffffc09bf3ee80
7204 [  680.566860] c0 7125 (syz-executor) page dumped because: VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page))
7205 [  680.578038] c0 7125 (syz-executor) page->mem_cgroup:ffffffc09bf3ee80
7206 [  680.585467] c0 7125 (syz-executor) ------------[ cut here ]------------
7207 [  680.592466] c0 7125 (syz-executor) kernel BUG at /home/build/farmland/adroid9.0/kernel/linux/mm/vmscan.c:1350!
7223 [  680.603663] c0 7125 (syz-executor) Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
7224 [  680.611436] c0 7125 (syz-executor) Modules linked in:
7225 [  680.616769] c0 7125 (syz-executor) CPU: 0 PID: 7125 Comm: syz-executor Tainted: G S              4.14.81 #3
7226 [  680.626826] c0 7125 (syz-executor) Hardware name: ASR AQUILAC EVB (DT)
7227 [  680.633623] c0 7125 (syz-executor) task: ffffffc00a54cd00 task.stack: ffffffc009b00000
7228 [  680.641917] c0 7125 (syz-executor) PC is at shrink_page_list+0x1998/0x3240
7229 [  680.649144] c0 7125 (syz-executor) LR is at shrink_page_list+0x1998/0x3240
7230 [  680.656303] c0 7125 (syz-executor) pc : [<ffffff90083a2158>] lr : [<ffffff90083a2158>] pstate: 60400045
7231 [  680.666086] c0 7125 (syz-executor) sp : ffffffc009b05940
..
7342 [  681.671308] c0 7125 (syz-executor) [<ffffff90083a2158>] shrink_page_list+0x1998/0x3240
7343 [  681.679567] c0 7125 (syz-executor) [<ffffff90083a3dc0>] reclaim_clean_pages_from_list+0x3c0/0x4f0
7344 [  681.688793] c0 7125 (syz-executor) [<ffffff900837ed64>] alloc_contig_range+0x3bc/0x650
7347 [  681.717421] c0 7125 (syz-executor) [<ffffff90084925cc>] cma_alloc+0x214/0x668
7348 [  681.724892] c0 7125 (syz-executor) [<ffffff90091e4d78>] ion_cma_allocate+0x98/0x1d8
7349 [  681.732872] c0 7125 (syz-executor) [<ffffff90091e0b20>] ion_alloc+0x200/0x7e0
7350 [  681.740302] c0 7125 (syz-executor) [<ffffff90091e154c>] ion_ioctl+0x18c/0x378
7351 [  681.747738] c0 7125 (syz-executor) [<ffffff90084c6824>] do_vfs_ioctl+0x17c/0x1780
7352 [  681.755514] c0 7125 (syz-executor) [<ffffff90084c7ed4>] SyS_ioctl+0xac/0xc0

Wu found it's due to [1]. Before that, unevictable page goes to cull_mlocked
routine so that it couldn't reach the VM_BUG_ON_PAGE line.

To fix the issue, this patch filter out unevictable LRU pages
from the reclaim_clean_pages_from_list in CMA.

[1] ad6b67041a45, mm: remove SWAP_MLOCK in ttu

Cc: <stable@xxxxxxxxxx>	[4.12+]
Reported-debugged-by: Wu Fangsuo <fangsuowu@xxxxxxxxxxxx>
Signed-off-by: Minchan Kim <minchan@xxxxxxxxxx>
---
 mm/vmscan.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index d9c3e873eca6..7350afae5c3c 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -1505,7 +1505,7 @@ unsigned long reclaim_clean_pages_from_list(struct zone *zone,
 
 	list_for_each_entry_safe(page, next, page_list, lru) {
 		if (page_is_file_cache(page) && !PageDirty(page) &&
-		    !__PageMovable(page)) {
+		    !__PageMovable(page) && !PageUnevictable(page)) {
 			ClearPageActive(page);
 			list_move(&page->lru, &clean_pages);
 		}
-- 
2.22.0.rc2.383.gf4fbbf30c2-goog





[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux