lkml.org 
[lkml]   [2018]   [Nov]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: Memory hotplug softlock issue
On Mon 19-11-18 13:51:21, Michal Hocko wrote:
> On Mon 19-11-18 13:40:33, Michal Hocko wrote:
> > On Mon 19-11-18 18:52:02, Baoquan He wrote:
> > [...]
> >
> > There are few stacks directly in the offline path but those should be
> > OK.
> > The real culprit seems to be the swap in code
> >
> > > [ +1.734416] CPU: 255 PID: 5558 Comm: stress Tainted: G L 4.20.0-rc2+ #7
> > > [ +0.007927] Hardware name: 9008/IT91SMUB, BIOS BLXSV512 03/22/2018
> > > [ +0.006297] Call Trace:
> > > [ +0.002537] dump_stack+0x46/0x60
> > > [ +0.003386] __migration_entry_wait.cold.65+0x5/0x14
> > > [ +0.005043] do_swap_page+0x84e/0x960
> > > [ +0.003727] ? arch_tlb_finish_mmu+0x29/0xc0
> > > [ +0.006412] __handle_mm_fault+0x933/0x1330
> > > [ +0.004265] handle_mm_fault+0xc4/0x250
> > > [ +0.003915] __do_page_fault+0x2b7/0x510
> > > [ +0.003990] do_page_fault+0x2c/0x110
> > > [ +0.003729] ? page_fault+0x8/0x30
> > > [ +0.003462] page_fault+0x1e/0x30
> >
> > There are many traces to this path. We are
> > /*
> > * Once page cache replacement of page migration started, page_count
> > * *must* be zero. And, we don't want to call wait_on_page_locked()
> > * against a page without get_page().
> > * So, we use get_page_unless_zero(), here. Even failed, page fault
> > * will occur again.
> > */
> > if (!get_page_unless_zero(page))
> > goto out;
> > pte_unmap_unlock(ptep, ptl);
> > wait_on_page_locked(page);
> >
> > taking a reference to the page under the migration. I have to think
> > about this much more but I suspec this is just calling for a problem.
> >
> > Cc migration experts. For you background information. We are seeing
> > memory offline not being able to converge because few heavily used pages
> > fail to migrate away - e.g. http://lkml.kernel.org/r/20181116012433.GU2653@MiWiFi-R3L-srv
> > A debugging page to dump stack for these pages http://lkml.kernel.org/r/20181116091409.GD14706@dhcp22.suse.cz
> > shows that references are taken from the swap in code (above). How are
> > we supposed to converge when the swapin code waits for the migration to
> > finish with the reference count elevated?
>
> Just to clarify. This is not only about swapin obviously. Any caller of
> __migration_entry_wait is affected the same way AFAICS.

In other words. Why cannot we do the following?

diff --git a/mm/migrate.c b/mm/migrate.c
index f7e4bfdc13b7..7ccab29bcf9a 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -324,19 +324,9 @@ void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
goto out;

page = migration_entry_to_page(entry);
-
- /*
- * Once page cache replacement of page migration started, page_count
- * *must* be zero. And, we don't want to call wait_on_page_locked()
- * against a page without get_page().
- * So, we use get_page_unless_zero(), here. Even failed, page fault
- * will occur again.
- */
- if (!get_page_unless_zero(page))
- goto out;
pte_unmap_unlock(ptep, ptl);
- wait_on_page_locked(page);
- put_page(page);
+ page_lock(page);
+ page_unlock(page);
return;
out:
pte_unmap_unlock(ptep, ptl);
--
Michal Hocko
SUSE Labs
\
 
 \ /
  Last update: 2018-11-19 15:11    [W:0.092 / U:13.808 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site