[PATCH] undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch

[PATCH] undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch

[Rik van Riel]

Undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch.
The brd pages do not live on the LRU at all, so there is no need to
play these tricks.

Debugged-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Rik van Riel <riel@redhat.com>

---
 drivers/block/brd.c |   13 -------------
 1 file changed, 13 deletions(-)

Index: linux-2.6.26-rc5-mm2/drivers/block/brd.c
===================================================================
--- linux-2.6.26-rc5-mm2.orig/drivers/block/brd.c	2008-06-11 14:39:43.000000000 -0400
+++ linux-2.6.26-rc5-mm2/drivers/block/brd.c	2008-06-12 13:49:18.000000000 -0400
@@ -374,21 +374,8 @@ static int brd_ioctl(struct inode *inode
 	return error;
 }
 
-/*
- * brd_open():
- * Just mark the mapping as containing unevictable pages
- */
-static int brd_open(struct inode *inode, struct file *filp)
-{
-	struct address_space *mapping = inode->i_mapping;
-
-	mapping_set_unevictable(mapping);
-	return 0;
-}
-
 static struct block_device_operations brd_fops = {
 	.owner =		THIS_MODULE,
-	.open  =		brd_open,
 	.ioctl =		brd_ioctl,
 #ifdef CONFIG_BLK_DEV_XIP
 	.direct_access =	brd_direct_access,

Re: [PATCH] undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch

[Lee Schermerhorn]

On Thu, 2008-06-12 at 13:53 -0400, Rik van Riel wrote:
> Undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch.
> The brd pages do not live on the LRU at all, so there is no need to
> play these tricks.

Does this mean that these pages cannot be migrated?  E.g., to evacuate
memory for hotplug?  Looks like all paths to migrate_pages() construct a
list of pages to migrate by isolating them from the lru
[isolate_lru_page()].  Any pages not found in the lru are skipped.

Lee
> 
> Debugged-by: Nick Piggin <nickpiggin@yahoo.com.au>
> Signed-off-by: Rik van Riel <riel@redhat.com>
> 
> ---
>  drivers/block/brd.c |   13 -------------
>  1 file changed, 13 deletions(-)
> 
> Index: linux-2.6.26-rc5-mm2/drivers/block/brd.c
> ===================================================================
> --- linux-2.6.26-rc5-mm2.orig/drivers/block/brd.c	2008-06-11 14:39:43.000000000 -0400
> +++ linux-2.6.26-rc5-mm2/drivers/block/brd.c	2008-06-12 13:49:18.000000000 -0400
> @@ -374,21 +374,8 @@ static int brd_ioctl(struct inode *inode
>  	return error;
>  }
>  
> -/*
> - * brd_open():
> - * Just mark the mapping as containing unevictable pages
> - */
> -static int brd_open(struct inode *inode, struct file *filp)
> -{
> -	struct address_space *mapping = inode->i_mapping;
> -
> -	mapping_set_unevictable(mapping);
> -	return 0;
> -}
> -
>  static struct block_device_operations brd_fops = {
>  	.owner =		THIS_MODULE,
> -	.open  =		brd_open,
>  	.ioctl =		brd_ioctl,
>  #ifdef CONFIG_BLK_DEV_XIP
>  	.direct_access =	brd_direct_access,
> 

Re: [PATCH] undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch

[Rik van Riel]

On Thu, 12 Jun 2008 15:00:57 -0400
Lee Schermerhorn <Lee.Schermerhorn@hp.com> wrote:

> On Thu, 2008-06-12 at 13:53 -0400, Rik van Riel wrote:
> > Undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch.
> > The brd pages do not live on the LRU at all, so there is no need to
> > play these tricks.
> 
> Does this mean that these pages cannot be migrated?  E.g., to evacuate
> memory for hotplug?  Looks like all paths to migrate_pages() construct a
> list of pages to migrate by isolating them from the lru
> [isolate_lru_page()].  Any pages not found in the lru are skipped.

It looks like that, indeed.

-- 
All Rights Reversed

Re: [PATCH] undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch

[Nick Piggin]

On Friday 13 June 2008 05:00, Lee Schermerhorn wrote:
> On Thu, 2008-06-12 at 13:53 -0400, Rik van Riel wrote:
> > Undo the brd.c part of ramfs-and-ram-disk-pages-are-unevictable.patch.
> > The brd pages do not live on the LRU at all, so there is no need to
> > play these tricks.
>
> Does this mean that these pages cannot be migrated?  E.g., to evacuate
> memory for hotplug?  Looks like all paths to migrate_pages() construct a
> list of pages to migrate by isolating them from the lru
> [isolate_lru_page()].  Any pages not found in the lru are skipped.

Yes. They have nothing to do with user pages or pagecache pages at all
really. brd is now a properly layered block device driver that does not
know anything about buffer cache ;)

You can think of the pages it allocates as completely private, and they
are used to implement the backing store for the block device.

The vm/fs layers will instantiate buffer cache over this device, which
is reclaimable.