2008-02-05 21:06:20

by Oliver Pinter

[permalink] [raw]
Subject: {2.6.22.y} quicklists must keep even off node pages on the quicklists until the TLB flush has been completed.

it is already im queue for 2.6.23,

----8<-----
>From [email protected] Sat Dec 22 14:04:08 2007
From: Christoph Lameter <[email protected]>
Date: Sat, 22 Dec 2007 14:03:23 -0800
Subject: quicklists: do not release off node pages early
To: [email protected]
Cc: [email protected], [email protected],
[email protected], [email protected]
Message-ID: <[email protected]>


From: Christoph Lameter <[email protected]>

patch ed367fc3a7349b17354c7acef551533337764859 in mainline.

quicklists must keep even off node pages on the quicklists until the TLB
flush has been completed.

Signed-off-by: Christoph Lameter <[email protected]>
Cc: Dhaval Giani <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>

---
include/linux/quicklist.h | 8 --------
1 file changed, 8 deletions(-)

--- a/include/linux/quicklist.h
+++ b/include/linux/quicklist.h
@@ -56,14 +56,6 @@ static inline void __quicklist_free(int
struct page *page)
{
struct quicklist *q;
- int nid = page_to_nid(page);
-
- if (unlikely(nid != numa_node_id())) {
- if (dtor)
- dtor(p);
- __free_page(page);
- return;
- }

q = &get_cpu_var(quicklist)[nr];
*(void **)p = q->page;

---->8------
Tested-by: Oliver Pinter <[email protected]> (on i386)

--
Thanks,
Oliver


2008-02-06 06:03:25

by Dhaval Giani

[permalink] [raw]
Subject: Re: {2.6.22.y} quicklists must keep even off node pages on the quicklists until the TLB flush has been completed.

On Tue, Feb 05, 2008 at 10:06:02PM +0100, Oliver Pinter wrote:
> it is already im queue for 2.6.23,
>
> ----8<-----
> >From [email protected] Sat Dec 22 14:04:08 2007
> From: Christoph Lameter <[email protected]>
> Date: Sat, 22 Dec 2007 14:03:23 -0800
> Subject: quicklists: do not release off node pages early
> To: [email protected]
> Cc: [email protected], [email protected],
> [email protected], [email protected]
> Message-ID: <[email protected]>
>
>
> From: Christoph Lameter <[email protected]>
>
> patch ed367fc3a7349b17354c7acef551533337764859 in mainline.
>
> quicklists must keep even off node pages on the quicklists until the TLB
> flush has been completed.
>
> Signed-off-by: Christoph Lameter <[email protected]>
> Cc: Dhaval Giani <[email protected]>
> Signed-off-by: Andrew Morton <[email protected]>
> Signed-off-by: Linus Torvalds <[email protected]>
> Signed-off-by: Greg Kroah-Hartman <[email protected]>
>
> ---
> include/linux/quicklist.h | 8 --------
> 1 file changed, 8 deletions(-)
>
> --- a/include/linux/quicklist.h
> +++ b/include/linux/quicklist.h
> @@ -56,14 +56,6 @@ static inline void __quicklist_free(int
> struct page *page)
> {
> struct quicklist *q;
> - int nid = page_to_nid(page);
> -
> - if (unlikely(nid != numa_node_id())) {
> - if (dtor)
> - dtor(p);
> - __free_page(page);
> - return;
> - }
>
> q = &get_cpu_var(quicklist)[nr];
> *(void **)p = q->page;
>
> ---->8------
> Tested-by: Oliver Pinter <[email protected]> (on i386)
>

Christoph,

Is this one also supposed to be backported?

--
regards,
Dhaval

2008-02-06 06:30:35

by Oliver Pinter

[permalink] [raw]
Subject: Re: {2.6.22.y} quicklists must keep even off node pages on the quicklists until the TLB flush has been completed.

I use this, without errors ... but the machine is i386 desktop

On Feb 6, 2008 7:02 AM, Dhaval Giani <[email protected]> wrote:
>
> On Tue, Feb 05, 2008 at 10:06:02PM +0100, Oliver Pinter wrote:
> > it is already im queue for 2.6.23,
> >
> > ----8<-----
> > >From [email protected] Sat Dec 22 14:04:08 2007
> > From: Christoph Lameter <[email protected]>
> > Date: Sat, 22 Dec 2007 14:03:23 -0800
> > Subject: quicklists: do not release off node pages early
> > To: [email protected]
> > Cc: [email protected], [email protected],
> > [email protected], [email protected]
> > Message-ID: <[email protected]>
> >
> >
> > From: Christoph Lameter <[email protected]>
> >
> > patch ed367fc3a7349b17354c7acef551533337764859 in mainline.
> >
> > quicklists must keep even off node pages on the quicklists until the TLB
> > flush has been completed.
> >
> > Signed-off-by: Christoph Lameter <[email protected]>
> > Cc: Dhaval Giani <[email protected]>
> > Signed-off-by: Andrew Morton <[email protected]>
> > Signed-off-by: Linus Torvalds <[email protected]>
> > Signed-off-by: Greg Kroah-Hartman <[email protected]>
> >
> > ---
> > include/linux/quicklist.h | 8 --------
> > 1 file changed, 8 deletions(-)
> >
> > --- a/include/linux/quicklist.h
> > +++ b/include/linux/quicklist.h
> > @@ -56,14 +56,6 @@ static inline void __quicklist_free(int
> > struct page *page)
> > {
> > struct quicklist *q;
> > - int nid = page_to_nid(page);
> > -
> > - if (unlikely(nid != numa_node_id())) {
> > - if (dtor)
> > - dtor(p);
> > - __free_page(page);
> > - return;
> > - }
> >
> > q = &get_cpu_var(quicklist)[nr];
> > *(void **)p = q->page;
> >
> > ---->8------
> > Tested-by: Oliver Pinter <[email protected]> (on i386)
> >
>
> Christoph,
>
> Is this one also supposed to be backported?
>
> --
> regards,
> Dhaval
>



--
Thanks,
Oliver

2008-02-06 18:58:24

by Christoph Lameter

[permalink] [raw]
Subject: Re: {2.6.22.y} quicklists must keep even off node pages on the quicklists until the TLB flush has been completed.

On Wed, 6 Feb 2008, Dhaval Giani wrote:

> Is this one also supposed to be backported?

Yes.

2008-02-06 18:58:58

by Christoph Lameter

[permalink] [raw]
Subject: Re: {2.6.22.y} quicklists must keep even off node pages on the quicklists until the TLB flush has been completed.

On Wed, 6 Feb 2008, Oliver Pinter wrote:

> I use this, without errors ... but the machine is i386 desktop

The fix for off node frees only applies to NUMA systems. !NUMA has no off
node pages.

2008-02-06 19:07:17

by Andrew Morton

[permalink] [raw]
Subject: Re: {2.6.22.y} quicklists must keep even off node pages on the quicklists until the TLB flush has been completed.

On Wed, 6 Feb 2008 10:58:13 -0800 (PST) Christoph Lameter <[email protected]> wrote:

> On Wed, 6 Feb 2008, Dhaval Giani wrote:
>
> > Is this one also supposed to be backported?
>
> Yes.

So mainline's ed367fc3a7349b17354c7acef551533337764859 needs to be backported
to 2.6.23.x.