Since commit bbddabe2e436aa7869b3ac5248df5c14ddde0cbf ("mm: filemap:
only do access activations on reads"), mark_page_accessed() is called
for reads only. But the idle flag is cleared by mark_page_accessed() so
the idle flag won't get cleared if the page is write accessed only.
Basically idle page tracking is used to estimate workingset size of
workload, noticeable size of workingset might be missed if the idle flag
is not maintained correctly.
It seems good enough to just clear idle flag for write operations.
Fixes: bbddabe2e436 ("mm: filemap: only do access activations on reads")
Cc: Johannes Weiner <[email protected]>
Cc: Rik van Riel <[email protected]>
Cc: Shakeel Butt <[email protected]>
Reported-by: Gang Deng <[email protected]>
Signed-off-by: Yang Shi <[email protected]>
---
mm/filemap.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/mm/filemap.c b/mm/filemap.c
index f0ae9a6..0589aef 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -41,6 +41,7 @@
#include <linux/delayacct.h>
#include <linux/psi.h>
#include <linux/ramfs.h>
+#include <linux/page_idle.h>
#include "internal.h"
#define CREATE_TRACE_POINTS
@@ -1630,6 +1631,11 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
if (fgp_flags & FGP_ACCESSED)
mark_page_accessed(page);
+ else if (fgp_flags & FGP_WRITE) {
+ /* Clear idle flag for buffer write */
+ if (page_is_idle(page))
+ clear_page_idle(page);
+ }
no_page:
if (!page && (fgp_flags & FGP_CREAT)) {
--
1.8.3.1
On Thu, 25 Jun 2020 01:43:32 +0800 Yang Shi <[email protected]> wrote:
> Since commit bbddabe2e436aa7869b3ac5248df5c14ddde0cbf ("mm: filemap:
> only do access activations on reads"), mark_page_accessed() is called
> for reads only. But the idle flag is cleared by mark_page_accessed() so
> the idle flag won't get cleared if the page is write accessed only.
>
> Basically idle page tracking is used to estimate workingset size of
> workload, noticeable size of workingset might be missed if the idle flag
> is not maintained correctly.
>
> It seems good enough to just clear idle flag for write operations.
>
> ...
>
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -41,6 +41,7 @@
> #include <linux/delayacct.h>
> #include <linux/psi.h>
> #include <linux/ramfs.h>
> +#include <linux/page_idle.h>
> #include "internal.h"
>
> #define CREATE_TRACE_POINTS
> @@ -1630,6 +1631,11 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
>
> if (fgp_flags & FGP_ACCESSED)
> mark_page_accessed(page);
> + else if (fgp_flags & FGP_WRITE) {
> + /* Clear idle flag for buffer write */
> + if (page_is_idle(page))
> + clear_page_idle(page);
> + }
>
> no_page:
> if (!page && (fgp_flags & FGP_CREAT)) {
The kerneldoc comment for pagecache_get_page() could do with some
updating - it fails to mention FGP_WRITE, FGP_NOFS and FGP_NOWAIT.
This change seems correct but also will have runtime effects. What are
they?
On 6/24/20 11:53 AM, Andrew Morton wrote:
> On Thu, 25 Jun 2020 01:43:32 +0800 Yang Shi <[email protected]> wrote:
>
>> Since commit bbddabe2e436aa7869b3ac5248df5c14ddde0cbf ("mm: filemap:
>> only do access activations on reads"), mark_page_accessed() is called
>> for reads only. But the idle flag is cleared by mark_page_accessed() so
>> the idle flag won't get cleared if the page is write accessed only.
>>
>> Basically idle page tracking is used to estimate workingset size of
>> workload, noticeable size of workingset might be missed if the idle flag
>> is not maintained correctly.
>>
>> It seems good enough to just clear idle flag for write operations.
>>
>> ...
>>
>> --- a/mm/filemap.c
>> +++ b/mm/filemap.c
>> @@ -41,6 +41,7 @@
>> #include <linux/delayacct.h>
>> #include <linux/psi.h>
>> #include <linux/ramfs.h>
>> +#include <linux/page_idle.h>
>> #include "internal.h"
>>
>> #define CREATE_TRACE_POINTS
>> @@ -1630,6 +1631,11 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
>>
>> if (fgp_flags & FGP_ACCESSED)
>> mark_page_accessed(page);
>> + else if (fgp_flags & FGP_WRITE) {
>> + /* Clear idle flag for buffer write */
>> + if (page_is_idle(page))
>> + clear_page_idle(page);
>> + }
>>
>> no_page:
>> if (!page && (fgp_flags & FGP_CREAT)) {
> The kerneldoc comment for pagecache_get_page() could do with some
> updating - it fails to mention FGP_WRITE, FGP_NOFS and FGP_NOWAIT.
Yes, will propose a separate patch later on.
>
> This change seems correct but also will have runtime effects. What are
> they?
Other than a couple of extra cycles when idle page tracking is enabled,
I didn't think of other effects. It should be negligible. The idle flag
doesn't play a role in page reclaim algorithm, so it won't have impact
on that.
On Wed, Jun 24, 2020 at 10:43 AM Yang Shi <[email protected]> wrote:
>
> Since commit bbddabe2e436aa7869b3ac5248df5c14ddde0cbf ("mm: filemap:
> only do access activations on reads"), mark_page_accessed() is called
> for reads only. But the idle flag is cleared by mark_page_accessed() so
> the idle flag won't get cleared if the page is write accessed only.
>
> Basically idle page tracking is used to estimate workingset size of
> workload, noticeable size of workingset might be missed if the idle flag
> is not maintained correctly.
>
> It seems good enough to just clear idle flag for write operations.
>
> Fixes: bbddabe2e436 ("mm: filemap: only do access activations on reads")
> Cc: Johannes Weiner <[email protected]>
> Cc: Rik van Riel <[email protected]>
> Cc: Shakeel Butt <[email protected]>
> Reported-by: Gang Deng <[email protected]>
> Signed-off-by: Yang Shi <[email protected]>
Reviewed-by: Shakeel Butt <[email protected]>
On Wed, Jun 24, 2020 at 12:18 PM Yang Shi <[email protected]> wrote:
>
>
>
> On 6/24/20 11:53 AM, Andrew Morton wrote:
> > On Thu, 25 Jun 2020 01:43:32 +0800 Yang Shi <[email protected]> wrote:
> >
> >> Since commit bbddabe2e436aa7869b3ac5248df5c14ddde0cbf ("mm: filemap:
> >> only do access activations on reads"), mark_page_accessed() is called
> >> for reads only. But the idle flag is cleared by mark_page_accessed() so
> >> the idle flag won't get cleared if the page is write accessed only.
> >>
> >> Basically idle page tracking is used to estimate workingset size of
> >> workload, noticeable size of workingset might be missed if the idle flag
> >> is not maintained correctly.
> >>
> >> It seems good enough to just clear idle flag for write operations.
> >>
> >> ...
> >>
> >> --- a/mm/filemap.c
> >> +++ b/mm/filemap.c
> >> @@ -41,6 +41,7 @@
> >> #include <linux/delayacct.h>
> >> #include <linux/psi.h>
> >> #include <linux/ramfs.h>
> >> +#include <linux/page_idle.h>
> >> #include "internal.h"
> >>
> >> #define CREATE_TRACE_POINTS
> >> @@ -1630,6 +1631,11 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
> >>
> >> if (fgp_flags & FGP_ACCESSED)
> >> mark_page_accessed(page);
> >> + else if (fgp_flags & FGP_WRITE) {
> >> + /* Clear idle flag for buffer write */
> >> + if (page_is_idle(page))
> >> + clear_page_idle(page);
> >> + }
> >>
> >> no_page:
> >> if (!page && (fgp_flags & FGP_CREAT)) {
> > The kerneldoc comment for pagecache_get_page() could do with some
> > updating - it fails to mention FGP_WRITE, FGP_NOFS and FGP_NOWAIT.
>
> Yes, will propose a separate patch later on.
>
> >
> > This change seems correct but also will have runtime effects. What are
> > they?
>
> Other than a couple of extra cycles when idle page tracking is enabled,
> I didn't think of other effects. It should be negligible. The idle flag
> doesn't play a role in page reclaim algorithm, so it won't have impact
> on that.
>
>
The only user visible impact will be on idle page tracking users. They
will get more accurate data.
On 6/24/20 12:50 PM, Shakeel Butt wrote:
> On Wed, Jun 24, 2020 at 12:18 PM Yang Shi <[email protected]> wrote:
>>
>>
>> On 6/24/20 11:53 AM, Andrew Morton wrote:
>>> On Thu, 25 Jun 2020 01:43:32 +0800 Yang Shi <[email protected]> wrote:
>>>
>>>> Since commit bbddabe2e436aa7869b3ac5248df5c14ddde0cbf ("mm: filemap:
>>>> only do access activations on reads"), mark_page_accessed() is called
>>>> for reads only. But the idle flag is cleared by mark_page_accessed() so
>>>> the idle flag won't get cleared if the page is write accessed only.
>>>>
>>>> Basically idle page tracking is used to estimate workingset size of
>>>> workload, noticeable size of workingset might be missed if the idle flag
>>>> is not maintained correctly.
>>>>
>>>> It seems good enough to just clear idle flag for write operations.
>>>>
>>>> ...
>>>>
>>>> --- a/mm/filemap.c
>>>> +++ b/mm/filemap.c
>>>> @@ -41,6 +41,7 @@
>>>> #include <linux/delayacct.h>
>>>> #include <linux/psi.h>
>>>> #include <linux/ramfs.h>
>>>> +#include <linux/page_idle.h>
>>>> #include "internal.h"
>>>>
>>>> #define CREATE_TRACE_POINTS
>>>> @@ -1630,6 +1631,11 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
>>>>
>>>> if (fgp_flags & FGP_ACCESSED)
>>>> mark_page_accessed(page);
>>>> + else if (fgp_flags & FGP_WRITE) {
>>>> + /* Clear idle flag for buffer write */
>>>> + if (page_is_idle(page))
>>>> + clear_page_idle(page);
>>>> + }
>>>>
>>>> no_page:
>>>> if (!page && (fgp_flags & FGP_CREAT)) {
>>> The kerneldoc comment for pagecache_get_page() could do with some
>>> updating - it fails to mention FGP_WRITE, FGP_NOFS and FGP_NOWAIT.
>> Yes, will propose a separate patch later on.
>>
>>> This change seems correct but also will have runtime effects. What are
>>> they?
>> Other than a couple of extra cycles when idle page tracking is enabled,
>> I didn't think of other effects. It should be negligible. The idle flag
>> doesn't play a role in page reclaim algorithm, so it won't have impact
>> on that.
>>
>>
> The only user visible impact will be on idle page tracking users. They
> will get more accurate data.
Thanks for elaborating this.