Skip to contentSkip to conversation

MADV_FREE support

PATCH Minchan Kim 2 participants 3 messages Feb 5, 2016 – Feb 7, 2016

Subsystems (2)
lore
Share & tools
lore
#0 lore raw
Quoted 20 lines · Michael Kerrisk (man-pages)

Hello Minchan,

Quoted 9 lines · Minchan Kim

In v4, Andrew wanted to settle in old basic MADV_FREE and introduces new stuffs(ie, lazyfree LRU, swapless support and lazyfreeness) later so this version doesn't include them.

I have been tested it on mmotm-2015-11-25-17-08 with additional patch[1] from Kirill to prevent BUG_ON which he didn't send to linux-mm yet as formal patch. With it, I couldn't find any problem so far.

Quoted 7 lines

Note that this version is based on THP refcount redesign so I needed some modification on MADV_FREE because split_huge_pmd doesn't split a THP page any more and pmd_trans_huge(pmd) is not enough to guarantee the page is not THP page.
As well, for MAVD_FREE lazy-split, THP split should respect pmd's dirtiness rather than marking ptes of all subpages dirty unconditionally. Please, review last patch in this patchset.

Quoted 5 lines

Now that MADV_FREE has been merged, would you be willing to write patch to the madvise(2) man page that describes the semantics, noes limitations and restrictions, and (ideally) has some sentences describing use cases?

Hello Michael,

Could you review this patch?

Thanks.

#1 lore raw

Hello Minchan,

Quoted 66 lines · Minchan Kim
Quoted 20 lines · Michael Kerrisk (man-pages)

Hello Minchan,

Quoted 9 lines · Minchan Kim

In v4, Andrew wanted to settle in old basic MADV_FREE and introduces new stuffs(ie, lazyfree LRU, swapless support and lazyfreeness) later so this version doesn't include them.

I have been tested it on mmotm-2015-11-25-17-08 with additional patch[1] from Kirill to prevent BUG_ON which he didn't send to linux-mm yet as formal patch. With it, I couldn't find any problem so far.

Quoted 7 lines

Note that this version is based on THP refcount redesign so I needed some modification on MADV_FREE because split_huge_pmd doesn't split a THP page any more and pmd_trans_huge(pmd) is not enough to guarantee the page is not THP page.
As well, for MAVD_FREE lazy-split, THP split should respect pmd's dirtiness rather than marking ptes of all subpages dirty unconditionally. Please, review last patch in this patchset.

Quoted 5 lines

Now that MADV_FREE has been merged, would you be willing to write patch to the madvise(2) man page that describes the semantics, noes limitations and restrictions, and (ideally) has some sentences describing use cases?

Hello Michael,

Could you review this patch?

Thanks.

Quoted 1 line

From 203372f901f574e991215fdff6907608ba53f932 Mon Sep 17 00:00:00 2001

From: Minchan Kim <[email protected]>
Date: Fri, 5 Feb 2016 11:09:54 +0900
Subject: [PATCH] madvise.2: Add MADV_FREE

Document the MADV_FREE flags added to madvise() in Linux 4.5

---
 man2/madvise.2 | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
man2/madvise.2 +12 -0
index c1df67c..4704304 100644
@@ -143,6 +143,25 @@ flag are special memory areas that are not managed
 by the virtual memory subsystem.
 Such pages are typically created by device drivers that
 map the pages into user space.)
+.TP
+.B MADV_FREE " (since Linux 4.5)"
+Application is finished with the given range, so kernel can free
+resources associated with it but the freeing could be delayed until
+memory pressure happens or canceld by write operation by user.
+
+After a successful MADV_FREE operation, user shouldn't expect kernel
+keeps stale data on the page. However, subsequent write of pages
+in the range will succeed and then kernel cannot free those dirtied pages
+so user can always see just written data. If there was no subsequent
+write, kernel can free those clean pages any time. In such case,
+user can see zero-fill-on-demand pages.

man2/madvise.2 flag are special memory areas that are not managed · line 167

Quoted 10 lines
man2/madvise.2 +7 -0
+
+Note that, it works only with private anonymous pages (see
+.BR mmap (2)).
+On swapless system, freeing pages in given range happens instantly
+regardless of memory pressure.
+
+
 .\"
 .\" ======================================================================
 .\"

Thanks for the nice text! I reworked somewhat, trying to fill out a few details about how I understand things work, but I may have introduced errors, so I would be happy if you would check the following text:

MADV_FREE (since Linux 4.5)
The application no longer requires the pages in the range specified by addr and len. The kernel can thus free these pages, but the freeing could be delayed until memory pressure occurs. For each of the pages that has been marked to be freed but has not yet been freed, the free operation will be canceled if the caller writes into the page. After a successful MADV_FREE operation, any stale data (i.e., dirty, unwritten pages) will be lost when the kernel frees the pages. However, subse‐ quent writes to pages in the range will succeed and then kernel cannot free those dirtied pages, so that the caller can always see just written data. If there is no subsequent write, the kernel can free the pages at any time. Once pages in the range have been freed, the caller will see zero-fill-on-demand pages upon subse‐ quent page references.

The MADV_FREE operation can be applied only to private anonymous pages (see mmap(2)). On a swapless system, freeing pages in a given range happens instantly, regardless of memory pressure.

Thanks,

Michael

Signature · 9 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#2 lore raw
Quoted 75 lines · Michael Kerrisk (man-pages)

Hello Minchan,

Quoted 72 lines · Minchan Kim
Quoted 20 lines · Michael Kerrisk (man-pages)

Hello Minchan,

Quoted 9 lines · Minchan Kim

In v4, Andrew wanted to settle in old basic MADV_FREE and introduces new stuffs(ie, lazyfree LRU, swapless support and lazyfreeness) later so this version doesn't include them.

I have been tested it on mmotm-2015-11-25-17-08 with additional patch[1] from Kirill to prevent BUG_ON which he didn't send to linux-mm yet as formal patch. With it, I couldn't find any problem so far.

Quoted 7 lines

Note that this version is based on THP refcount redesign so I needed some modification on MADV_FREE because split_huge_pmd doesn't split a THP page any more and pmd_trans_huge(pmd) is not enough to guarantee the page is not THP page.
As well, for MAVD_FREE lazy-split, THP split should respect pmd's dirtiness rather than marking ptes of all subpages dirty unconditionally. Please, review last patch in this patchset.

Quoted 5 lines

Now that MADV_FREE has been merged, would you be willing to write patch to the madvise(2) man page that describes the semantics, noes limitations and restrictions, and (ideally) has some sentences describing use cases?

Hello Michael,

Could you review this patch?

Thanks.

Quoted 1 line

From 203372f901f574e991215fdff6907608ba53f932 Mon Sep 17 00:00:00 2001

From: Minchan Kim <[email protected]>
Date: Fri, 5 Feb 2016 11:09:54 +0900
Subject: [PATCH] madvise.2: Add MADV_FREE

Document the MADV_FREE flags added to madvise() in Linux 4.5

---
 man2/madvise.2 | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
man2/madvise.2 +19 -0
index c1df67c..4704304 100644
@@ -143,6 +143,25 @@ flag are special memory areas that are not managed
 by the virtual memory subsystem.
 Such pages are typically created by device drivers that
 map the pages into user space.)
+.TP
+.B MADV_FREE " (since Linux 4.5)"
+Application is finished with the given range, so kernel can free
+resources associated with it but the freeing could be delayed until
+memory pressure happens or canceld by write operation by user.
+
+After a successful MADV_FREE operation, user shouldn't expect kernel
+keeps stale data on the page. However, subsequent write of pages
+in the range will succeed and then kernel cannot free those dirtied pages
+so user can always see just written data. If there was no subsequent
+write, kernel can free those clean pages any time. In such case,
+user can see zero-fill-on-demand pages.
+
+Note that, it works only with private anonymous pages (see
+.BR mmap (2)).
+On swapless system, freeing pages in given range happens instantly
+regardless of memory pressure.
+
+
Quoted 8 lines

man2/madvise.2 flag are special memory areas that are not managed · line 167

Quoted 4 lines
man2/madvise.2 +0 -0
 .\"
 .\" ======================================================================
 .\"

Thanks for the nice text! I reworked somewhat, trying to fill out a few details about how I understand things work, but I may have introduced errors, so I would be happy if you would check the following text:

Below looks good to me.
Thanks, Michael

Quoted 19 lines

MADV_FREE (since Linux 4.5)
The application no longer requires the pages in the range specified by addr and len. The kernel can thus free these pages, but the freeing could be delayed until memory pressure occurs. For each of the pages that has been marked to be freed but has not yet been freed, the free operation will be canceled if the caller writes into the page. After a successful MADV_FREE operation, any stale data (i.e., dirty, unwritten pages) will be lost when the kernel frees the pages. However, subse‐ quent writes to pages in the range will succeed and then kernel cannot free those dirtied pages, so that the caller can always see just written data. If there is no subsequent write, the kernel can free the pages at any time. Once pages in the range have been freed, the caller will see zero-fill-on-demand pages upon subse‐ quent page references.

Quoted 19 lines

The MADV_FREE operation can be applied only to private anonymous pages (see mmap(2)). On a swapless system, freeing pages in a given range happens instantly, regardless of memory pressure.

Thanks,

Michael

Signature · 9 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>