On Tue, Feb 06, 2018 at 09:34:44PM +0800, huang ying wrote:
> On Tue, Feb 6, 2018 at 5:02 PM, Minchan Kim <minchan@xxxxxxxxxx> wrote:
> > On Tue, Feb 06, 2018 at 04:39:18PM +0800, Huang, Ying wrote:
> >> Hi, Minchan,
> >>
> >> Minchan Kim <minchan@xxxxxxxxxx> writes:
> >>
> >> > Hi Huang,
> >> >
> >> > On Tue, Feb 06, 2018 at 02:54:04PM +0800, Huang, Ying wrote:
> >> >> From: Huang Ying <ying.huang@xxxxxxxxx>
> >> >>
> >> >> It was reported by Sergey Senozhatsky that if THP (Transparent Huge
> >> >> Page) and frontswap (via zswap) are both enabled, when memory goes low
> >> >> so that swap is triggered, segfault and memory corruption will occur
> >> >> in random user space applications as follow,
> >> >>
> >> >> kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]
> >> >> #0 0x00007fc08889ae0d _int_malloc (libc.so.6)
> >> >> #1 0x00007fc08889c2f3 malloc (libc.so.6)
> >> >> #2 0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
> >> >> #3 0x0000560e6005e75c n/a (urxvt)
> >> >> #4 0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
> >> >> #5 0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
> >> >> #6 0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
> >> >> #7 0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
> >> >> #8 0x0000560e6005cb55 ev_run (urxvt)
> >> >> #9 0x0000560e6003b9b9 main (urxvt)
> >> >> #10 0x00007fc08883af4a __libc_start_main (libc.so.6)
> >> >> #11 0x0000560e6003f9da _start (urxvt)
> >> >>
> >> >> After bisection, it was found the first bad commit is
> >> >> bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped
> >> >> out").
> >> >>
> >> >> The root cause is as follow.
> >> >>
> >> >> When the pages are written to storage device during swapping out in
> >> >> swap_writepage(), zswap (fontswap) is tried to compress the pages
> >> >> instead to improve the performance. But zswap (frontswap) will treat
> >> >> THP as normal page, so only the head page is saved. After swapping
> >> >> in, tail pages will not be restored to its original contents, so cause
> >> >> the memory corruption in the applications.
> >> >>
> >> >> This is fixed via splitting THP at the begin of swapping out if
> >> >> frontswap is enabled. To avoid frontswap to be enabled at runtime,
> >> >> whether the page is THP is checked before using frontswap during
> >> >> swapping out too.
> >> >
> >> > Nice catch, Huang. However, before the adding a new dependency between
> >> > frontswap and vmscan that I want to avoid if it is possible, let's think
> >> > whether frontswap can support THP page or not.
> >> > Can't we handle it with some loop to handle all of subpages of THP page?
> >> > It might be not hard?
> >>
> >> Yes. That could be an optimization over this patch. This patch is just
> >> a simple fix to make things work and be suitable for stable tree.
> >
> > Yub, it would be more complex than this patch. However, this patch introduces
> > a new dependency to vmscan.c. IOW, we have been good without knowing frontswap
> > in vmscan.c but from now on, we should be aware of that, which is unfortunate.
> >
> > Can't we simple do like that if you want to make it simple and rely on someone
> > who makes frontswap THP-aware later?
> >
> > diff --git a/mm/swapfile.c b/mm/swapfile.c
> > index 42fe5653814a..4bf1725407aa 100644
> > --- a/mm/swapfile.c
> > +++ b/mm/swapfile.c
> > @@ -934,7 +934,11 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
> >
> > /* Only single cluster request supported */
> > WARN_ON_ONCE(n_goal > 1 && cluster);
> > +#ifdef CONFIG_FRONTSWAP
> > + /* Now, frontswap doesn't support THP page */
> > + if (frontswap_enabled() && cluster)
> > + return;
> > +#endif
> > avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages;
> > if (avail_pgs <= 0)
> > goto noswap;
> >
>
> This can avoid introduce dependency on frontswap in vmscan.c. But
> IMHO it doesn't look like the right place to place the logic.
> vmscan.c is the place we put policy to determine whether to split THP.
It adds split policy in vmscan.c like you said.
shrink_page_list already relies on swap_file.c to decide split a THP page.
IOW, if a THP swap stuff is not avilable, split a thp.
It's totally same logic. I don't see any difference at all.
shrink_page_list:
if (!add_to_swap(page)) {
if (PageTransHuge(page))
goto activate_locked;
if (split_huge_page_to_list(page, page_list))
goto activate_locked;
count_vm_event(THP_SWPOUT_FALLBACK);
if (!add_to_swap(page))
goto activate_locked;
}