diff mbox series

mm: thp: clear PageDoubleMap flag when the last PMD map gone

Message ID 1571938066-29031-1-git-send-email-yang.shi@linux.alibaba.com (mailing list archive)
State New, archived
Headers show
Series mm: thp: clear PageDoubleMap flag when the last PMD map gone | expand

Commit Message

Yang Shi Oct. 24, 2019, 5:27 p.m. UTC
File THP sets PageDoubleMap flag when the first it gets PTE mapped, but
the flag is never cleared until the THP is freed.  This result in
unbalanced state although it is not a big deal. 

Clear the flag when the last compound_mapcount is gone.  It should be
cleared when all the PTE maps are gone (become PMD mapped only) as well,
but this needs check all subpage's _mapcount every time any subpage's
rmap is removed, the overhead may be not worth.  The anonymous THP also
just clears PageDoubleMap flag when the last PMD map is gone.

Cc: Hugh Dickins <hughd@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
---
Hugh thought it is unnecessary to fix it completely due to the overhead
(https://lkml.org/lkml/2019/10/22/1011), but it sounds simple to achieve
the similar balance as anonymous THP.

 mm/rmap.c | 3 +++
 1 file changed, 3 insertions(+)

Comments

Song Liu Oct. 24, 2019, 7:26 p.m. UTC | #1
On Thu, Oct 24, 2019 at 10:28 AM Yang Shi <yang.shi@linux.alibaba.com> wrote:
>
> File THP sets PageDoubleMap flag when the first it gets PTE mapped, but
> the flag is never cleared until the THP is freed.  This result in
> unbalanced state although it is not a big deal.
>
> Clear the flag when the last compound_mapcount is gone.  It should be
> cleared when all the PTE maps are gone (become PMD mapped only) as well,
> but this needs check all subpage's _mapcount every time any subpage's
> rmap is removed, the overhead may be not worth.  The anonymous THP also
> just clears PageDoubleMap flag when the last PMD map is gone.
>
> Cc: Hugh Dickins <hughd@google.com>
> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Cc: Andrea Arcangeli <aarcange@redhat.com>
> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>

Looks good to me. Thanks!

Acked-by: Song Liu <songliubraving@fb.com>

> ---
> Hugh thought it is unnecessary to fix it completely due to the overhead
> (https://lkml.org/lkml/2019/10/22/1011), but it sounds simple to achieve
> the similar balance as anonymous THP.
>
>  mm/rmap.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 0c7b2a9..d17cbf3 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1236,6 +1236,9 @@ static void page_remove_file_rmap(struct page *page, bool compound)
>                         __dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
>                 else
>                         __dec_node_page_state(page, NR_FILE_PMDMAPPED);
> +
> +               /* The last PMD map is gone */
> +               ClearPageDoubleMap(compound_head(page));
>         } else {
>                 if (!atomic_add_negative(-1, &page->_mapcount))
>                         goto out;
> --
> 1.8.3.1
>
>
Kirill A. Shutemov Oct. 25, 2019, 3:36 p.m. UTC | #2
On Fri, Oct 25, 2019 at 01:27:46AM +0800, Yang Shi wrote:
> File THP sets PageDoubleMap flag when the first it gets PTE mapped, but
> the flag is never cleared until the THP is freed.  This result in
> unbalanced state although it is not a big deal. 
> 
> Clear the flag when the last compound_mapcount is gone.  It should be
> cleared when all the PTE maps are gone (become PMD mapped only) as well,
> but this needs check all subpage's _mapcount every time any subpage's
> rmap is removed, the overhead may be not worth.  The anonymous THP also
> just clears PageDoubleMap flag when the last PMD map is gone.

NAK, sorry.

The key difference with anon THP that file THP can be mapped again with
PMD after all PMD (or all) mappings are gone.

Your patch breaks the case when you map the page with PMD again while the
page is still mapped with PTEs. Who would set PageDoubleMap() in this
case?
Yang Shi Oct. 25, 2019, 3:58 p.m. UTC | #3
On 10/25/19 8:36 AM, Kirill A. Shutemov wrote:
> On Fri, Oct 25, 2019 at 01:27:46AM +0800, Yang Shi wrote:
>> File THP sets PageDoubleMap flag when the first it gets PTE mapped, but
>> the flag is never cleared until the THP is freed.  This result in
>> unbalanced state although it is not a big deal.
>>
>> Clear the flag when the last compound_mapcount is gone.  It should be
>> cleared when all the PTE maps are gone (become PMD mapped only) as well,
>> but this needs check all subpage's _mapcount every time any subpage's
>> rmap is removed, the overhead may be not worth.  The anonymous THP also
>> just clears PageDoubleMap flag when the last PMD map is gone.
> NAK, sorry.
>
> The key difference with anon THP that file THP can be mapped again with
> PMD after all PMD (or all) mappings are gone.
>
> Your patch breaks the case when you map the page with PMD again while the
> page is still mapped with PTEs. Who would set PageDoubleMap() in this
> case?

Aha, yes, you are right. I missed that point. However, I'm wondering we 
might move this up a little bit like this:

diff --git a/mm/rmap.c b/mm/rmap.c
index d17cbf3..ac046fd 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1230,15 +1230,17 @@ static void page_remove_file_rmap(struct page 
*page, bool compound)
                         if (atomic_add_negative(-1, &page[i]._mapcount))
                                 nr++;
                 }
+
+               /* No PTE map anymore */
+               if (nr == HPAGE_PMD_NR)
+                       ClearPageDoubleMap(compound_head(page));
+
                 if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
                         goto out;
                 if (PageSwapBacked(page))
                         __dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
                 else
                         __dec_node_page_state(page, NR_FILE_PMDMAPPED);
-
-               /* The last PMD map is gone */
-               ClearPageDoubleMap(compound_head(page));
         } else {
                 if (!atomic_add_negative(-1, &page->_mapcount))
                         goto out;


This should guarantee no PTE map anymore, it should be safe to clear the 
flag.

>
Kirill A. Shutemov Oct. 25, 2019, 4:32 p.m. UTC | #4
On Fri, Oct 25, 2019 at 08:58:22AM -0700, Yang Shi wrote:
> 
> 
> On 10/25/19 8:36 AM, Kirill A. Shutemov wrote:
> > On Fri, Oct 25, 2019 at 01:27:46AM +0800, Yang Shi wrote:
> > > File THP sets PageDoubleMap flag when the first it gets PTE mapped, but
> > > the flag is never cleared until the THP is freed.  This result in
> > > unbalanced state although it is not a big deal.
> > > 
> > > Clear the flag when the last compound_mapcount is gone.  It should be
> > > cleared when all the PTE maps are gone (become PMD mapped only) as well,
> > > but this needs check all subpage's _mapcount every time any subpage's
> > > rmap is removed, the overhead may be not worth.  The anonymous THP also
> > > just clears PageDoubleMap flag when the last PMD map is gone.
> > NAK, sorry.
> > 
> > The key difference with anon THP that file THP can be mapped again with
> > PMD after all PMD (or all) mappings are gone.
> > 
> > Your patch breaks the case when you map the page with PMD again while the
> > page is still mapped with PTEs. Who would set PageDoubleMap() in this
> > case?
> 
> Aha, yes, you are right. I missed that point. However, I'm wondering we
> might move this up a little bit like this:
> 
> diff --git a/mm/rmap.c b/mm/rmap.c
> index d17cbf3..ac046fd 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1230,15 +1230,17 @@ static void page_remove_file_rmap(struct page *page,
> bool compound)
>                         if (atomic_add_negative(-1, &page[i]._mapcount))
>                                 nr++;
>                 }
> +
> +               /* No PTE map anymore */
> +               if (nr == HPAGE_PMD_NR)
> +                       ClearPageDoubleMap(compound_head(page));
> +
>                 if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
>                         goto out;
>                 if (PageSwapBacked(page))
>                         __dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
>                 else
>                         __dec_node_page_state(page, NR_FILE_PMDMAPPED);
> -
> -               /* The last PMD map is gone */
> -               ClearPageDoubleMap(compound_head(page));
>         } else {
>                 if (!atomic_add_negative(-1, &page->_mapcount))
>                         goto out;
> 
> 
> This should guarantee no PTE map anymore, it should be safe to clear the
> flag.

At first glance looks safe, but let me think more about it. I didn't
expect it be that easy :P
Kirill A. Shutemov Oct. 25, 2019, 4:39 p.m. UTC | #5
On Fri, Oct 25, 2019 at 07:32:33PM +0300, Kirill A. Shutemov wrote:
> On Fri, Oct 25, 2019 at 08:58:22AM -0700, Yang Shi wrote:
> > 
> > 
> > On 10/25/19 8:36 AM, Kirill A. Shutemov wrote:
> > > On Fri, Oct 25, 2019 at 01:27:46AM +0800, Yang Shi wrote:
> > > > File THP sets PageDoubleMap flag when the first it gets PTE mapped, but
> > > > the flag is never cleared until the THP is freed.  This result in
> > > > unbalanced state although it is not a big deal.
> > > > 
> > > > Clear the flag when the last compound_mapcount is gone.  It should be
> > > > cleared when all the PTE maps are gone (become PMD mapped only) as well,
> > > > but this needs check all subpage's _mapcount every time any subpage's
> > > > rmap is removed, the overhead may be not worth.  The anonymous THP also
> > > > just clears PageDoubleMap flag when the last PMD map is gone.
> > > NAK, sorry.
> > > 
> > > The key difference with anon THP that file THP can be mapped again with
> > > PMD after all PMD (or all) mappings are gone.
> > > 
> > > Your patch breaks the case when you map the page with PMD again while the
> > > page is still mapped with PTEs. Who would set PageDoubleMap() in this
> > > case?
> > 
> > Aha, yes, you are right. I missed that point. However, I'm wondering we
> > might move this up a little bit like this:
> > 
> > diff --git a/mm/rmap.c b/mm/rmap.c
> > index d17cbf3..ac046fd 100644
> > --- a/mm/rmap.c
> > +++ b/mm/rmap.c
> > @@ -1230,15 +1230,17 @@ static void page_remove_file_rmap(struct page *page,
> > bool compound)
> >                         if (atomic_add_negative(-1, &page[i]._mapcount))
> >                                 nr++;
> >                 }
> > +
> > +               /* No PTE map anymore */
> > +               if (nr == HPAGE_PMD_NR)
> > +                       ClearPageDoubleMap(compound_head(page));
> > +
> >                 if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
> >                         goto out;
> >                 if (PageSwapBacked(page))
> >                         __dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
> >                 else
> >                         __dec_node_page_state(page, NR_FILE_PMDMAPPED);
> > -
> > -               /* The last PMD map is gone */
> > -               ClearPageDoubleMap(compound_head(page));
> >         } else {
> >                 if (!atomic_add_negative(-1, &page->_mapcount))
> >                         goto out;
> > 
> > 
> > This should guarantee no PTE map anymore, it should be safe to clear the
> > flag.
> 
> At first glance looks safe, but let me think more about it. I didn't
> expect it be that easy :P

How do you protect from races? What prevents other thread/process to map
the page as PTE after you've calculated 'nr'?

I don't remember the code that well, but I believe we don't require
PageLock for all cases... Or do we?
Yang Shi Oct. 25, 2019, 6:49 p.m. UTC | #6
On 10/25/19 9:39 AM, Kirill A. Shutemov wrote:
> On Fri, Oct 25, 2019 at 07:32:33PM +0300, Kirill A. Shutemov wrote:
>> On Fri, Oct 25, 2019 at 08:58:22AM -0700, Yang Shi wrote:
>>>
>>> On 10/25/19 8:36 AM, Kirill A. Shutemov wrote:
>>>> On Fri, Oct 25, 2019 at 01:27:46AM +0800, Yang Shi wrote:
>>>>> File THP sets PageDoubleMap flag when the first it gets PTE mapped, but
>>>>> the flag is never cleared until the THP is freed.  This result in
>>>>> unbalanced state although it is not a big deal.
>>>>>
>>>>> Clear the flag when the last compound_mapcount is gone.  It should be
>>>>> cleared when all the PTE maps are gone (become PMD mapped only) as well,
>>>>> but this needs check all subpage's _mapcount every time any subpage's
>>>>> rmap is removed, the overhead may be not worth.  The anonymous THP also
>>>>> just clears PageDoubleMap flag when the last PMD map is gone.
>>>> NAK, sorry.
>>>>
>>>> The key difference with anon THP that file THP can be mapped again with
>>>> PMD after all PMD (or all) mappings are gone.
>>>>
>>>> Your patch breaks the case when you map the page with PMD again while the
>>>> page is still mapped with PTEs. Who would set PageDoubleMap() in this
>>>> case?
>>> Aha, yes, you are right. I missed that point. However, I'm wondering we
>>> might move this up a little bit like this:
>>>
>>> diff --git a/mm/rmap.c b/mm/rmap.c
>>> index d17cbf3..ac046fd 100644
>>> --- a/mm/rmap.c
>>> +++ b/mm/rmap.c
>>> @@ -1230,15 +1230,17 @@ static void page_remove_file_rmap(struct page *page,
>>> bool compound)
>>>                          if (atomic_add_negative(-1, &page[i]._mapcount))
>>>                                  nr++;
>>>                  }
>>> +
>>> +               /* No PTE map anymore */
>>> +               if (nr == HPAGE_PMD_NR)
>>> +                       ClearPageDoubleMap(compound_head(page));
>>> +
>>>                  if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
>>>                          goto out;
>>>                  if (PageSwapBacked(page))
>>>                          __dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
>>>                  else
>>>                          __dec_node_page_state(page, NR_FILE_PMDMAPPED);
>>> -
>>> -               /* The last PMD map is gone */
>>> -               ClearPageDoubleMap(compound_head(page));
>>>          } else {
>>>                  if (!atomic_add_negative(-1, &page->_mapcount))
>>>                          goto out;
>>>
>>>
>>> This should guarantee no PTE map anymore, it should be safe to clear the
>>> flag.
>> At first glance looks safe, but let me think more about it. I didn't
>> expect it be that easy :P
> How do you protect from races? What prevents other thread/process to map
> the page as PTE after you've calculated 'nr'?
>
> I don't remember the code that well, but I believe we don't require
> PageLock for all cases... Or do we?

No, page lock is required by adding PTE rmap, but not required when 
removing rmap, i.e. huge pmd split. It looks we can't prevent from the 
races for processes, threads are protected by ptl.

>
diff mbox series

Patch

diff --git a/mm/rmap.c b/mm/rmap.c
index 0c7b2a9..d17cbf3 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1236,6 +1236,9 @@  static void page_remove_file_rmap(struct page *page, bool compound)
 			__dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
 		else
 			__dec_node_page_state(page, NR_FILE_PMDMAPPED);
+
+		/* The last PMD map is gone */
+		ClearPageDoubleMap(compound_head(page));
 	} else {
 		if (!atomic_add_negative(-1, &page->_mapcount))
 			goto out;