2012-08-21 12:51:22

by Xishi Qiu

Subject: [PATCH] memory-hotplug: add build zonelists when offline pages

From: Xishi Qiu <[email protected]>

online_pages() does build_all_zonelists() and zone_pcp_update(),
I think offline_pages() should do it too. The node has no memory
to allocate, so remove this node's zones form other nodes' zonelists.


Signed-off-by: Xishi Qiu <[email protected]>
---
mm/memory_hotplug.c | 6 +++++-
1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index bc7e7a2..5172bd4 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -979,7 +979,11 @@ repeat:
if (!node_present_pages(node)) {
node_clear_state(node, N_HIGH_MEMORY);
kswapd_stop(node);
- }
+ mutex_lock(&zonelists_mutex);
+ build_all_zonelists(NODE_DATA(node), NULL);
+ mutex_unlock(&zonelists_mutex);
+ } else
+ zone_pcp_update(zone);

vm_total_pages = nr_free_pagecache_pages();
writeback_set_ratelimit();
--
1.7.6.1

2012-08-22 06:10:22

by Wen Congyang

Subject: Re: [PATCH] memory-hotplug: add build zonelists when offline pages

At 08/21/2012 08:51 PM, qiuxishi Wrote:
> From: Xishi Qiu <[email protected]>
>
> online_pages() does build_all_zonelists() and zone_pcp_update(),
> I think offline_pages() should do it too. The node has no memory
> to allocate, so remove this node's zones form other nodes' zonelists.
>
>
> Signed-off-by: Xishi Qiu <[email protected]>
> ---
> mm/memory_hotplug.c | 6 +++++-
> 1 files changed, 5 insertions(+), 1 deletions(-)
>
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index bc7e7a2..5172bd4 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -979,7 +979,11 @@ repeat:
> if (!node_present_pages(node)) {
> node_clear_state(node, N_HIGH_MEMORY);
> kswapd_stop(node);
> - }
> + mutex_lock(&zonelists_mutex);
> + build_all_zonelists(NODE_DATA(node), NULL);

The node is still onlined now, so there is no need to pass
this node's pgdat to build_all_zonelists().

I think we should build all zonelists when the zone has no
pages.

> + mutex_unlock(&zonelists_mutex);
> + } else
> + zone_pcp_update(zone);

There is more than one zone in a node. So the zone can have
no pages when the node has some pages.

And we have called drain_all_pages(), I think there is no need
to call zone_pcp_update() here.

Thanks
Wen Congyang

>
> vm_total_pages = nr_free_pagecache_pages();
> writeback_set_ratelimit();

2012-08-22 08:00:09

by Xishi Qiu

Subject: Re: [PATCH] memory-hotplug: add build zonelists when offline pages

On 2012-8-22 14:15, Wen Congyang wrote:
> At 08/21/2012 08:51 PM, qiuxishi Wrote:
>> From: Xishi Qiu <[email protected]>
>>
>> online_pages() does build_all_zonelists() and zone_pcp_update(),
>> I think offline_pages() should do it too. The node has no memory
>> to allocate, so remove this node's zones form other nodes' zonelists.
>>
>>
>> Signed-off-by: Xishi Qiu <[email protected]>
>> ---
>> mm/memory_hotplug.c | 6 +++++-
>> 1 files changed, 5 insertions(+), 1 deletions(-)
>>
>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>> index bc7e7a2..5172bd4 100644
>> --- a/mm/memory_hotplug.c
>> +++ b/mm/memory_hotplug.c
>> @@ -979,7 +979,11 @@ repeat:
>> if (!node_present_pages(node)) {
>> node_clear_state(node, N_HIGH_MEMORY);
>> kswapd_stop(node);
>> - }
>> + mutex_lock(&zonelists_mutex);
>> + build_all_zonelists(NODE_DATA(node), NULL);
>
> The node is still onlined now, so there is no need to pass
> this node's pgdat to build_all_zonelists().
>
> I think we should build all zonelists when the zone has no
> pages.
>
>> + mutex_unlock(&zonelists_mutex);
>> + } else
>> + zone_pcp_update(zone);
>
> There is more than one zone in a node. So the zone can have
> no pages when the node has some pages.
>

Yes, you are right. Here is the new patch,

Signed-off-by: Xishi Qiu <[email protected]>
---
mm/memory_hotplug.c | 7 ++++++-
1 files changed, 6 insertions(+), 1 deletions(-)

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index bc7e7a2..5f6997f 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -973,8 +973,13 @@ repeat:

init_per_zone_wmark_min();

- if (!populated_zone(zone))
+ if (!populated_zone(zone)) {
zone_pcp_reset(zone);
+ mutex_lock(&zonelists_mutex);
+ build_all_zonelists(NULL, NULL);
+ mutex_unlock(&zonelists_mutex);
+ } else
+ zone_pcp_update(zone);

if (!node_present_pages(node)) {
node_clear_state(node, N_HIGH_MEMORY);
--
1.7.6.1

> And we have called drain_all_pages(), I think there is no need
> to call zone_pcp_update() here.
>
> Thanks
> Wen Congyang
>

In zone_pcp_update(), it calculates zone_batchsize() which does
not calculated in drain_all_pages().

Thanks
Xishi Qiu

>>
>> vm_total_pages = nr_free_pagecache_pages();
>> writeback_set_ratelimit();
>
>

2012-08-22 08:56:31

by Wen Congyang

Subject: Re: [PATCH] memory-hotplug: add build zonelists when offline pages

At 08/22/2012 03:59 PM, qiuxishi Wrote:
> On 2012-8-22 14:15, Wen Congyang wrote:
>> > At 08/21/2012 08:51 PM, qiuxishi Wrote:
>>> >> From: Xishi Qiu <[email protected]>
>>> >>
>>> >> online_pages() does build_all_zonelists() and zone_pcp_update(),
>>> >> I think offline_pages() should do it too. The node has no memory
>>> >> to allocate, so remove this node's zones form other nodes' zonelists.
>>> >>
>>> >>
>>> >> Signed-off-by: Xishi Qiu <[email protected]>
>>> >> ---
>>> >> mm/memory_hotplug.c | 6 +++++-
>>> >> 1 files changed, 5 insertions(+), 1 deletions(-)
>>> >>
>>> >> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>>> >> index bc7e7a2..5172bd4 100644
>>> >> --- a/mm/memory_hotplug.c
>>> >> +++ b/mm/memory_hotplug.c
>>> >> @@ -979,7 +979,11 @@ repeat:
>>> >> if (!node_present_pages(node)) {
>>> >> node_clear_state(node, N_HIGH_MEMORY);
>>> >> kswapd_stop(node);
>>> >> - }
>>> >> + mutex_lock(&zonelists_mutex);
>>> >> + build_all_zonelists(NODE_DATA(node), NULL);
>> >
>> > The node is still onlined now, so there is no need to pass
>> > this node's pgdat to build_all_zonelists().
>> >
>> > I think we should build all zonelists when the zone has no
>> > pages.
>> >
>>> >> + mutex_unlock(&zonelists_mutex);
>>> >> + } else
>>> >> + zone_pcp_update(zone);
>> >
>> > There is more than one zone in a node. So the zone can have
>> > no pages when the node has some pages.
>> >
> Yes, you are right. Here is the new patch,

This version looks fine to me.

>
> Signed-off-by: Xishi Qiu <[email protected]>
> ---
> mm/memory_hotplug.c | 7 ++++++-
> 1 files changed, 6 insertions(+), 1 deletions(-)
>
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index bc7e7a2..5f6997f 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -973,8 +973,13 @@ repeat:
>
> init_per_zone_wmark_min();
>
> - if (!populated_zone(zone))
> + if (!populated_zone(zone)) {
> zone_pcp_reset(zone);
> + mutex_lock(&zonelists_mutex);
> + build_all_zonelists(NULL, NULL);
> + mutex_unlock(&zonelists_mutex);
> + } else
> + zone_pcp_update(zone);
>
> if (!node_present_pages(node)) {
> node_clear_state(node, N_HIGH_MEMORY);
> -- 1.7.6.1
>> > And we have called drain_all_pages(), I think there is no need
>> > to call zone_pcp_update() here.
>> >
>> > Thanks
>> > Wen Congyang
>> >
> In zone_pcp_update(), it calculates zone_batchsize() which does
> not calculated in drain_all_pages().

Yes, zone_batchsize() depends on zone's present pages.
If zone's present pages are changed, zone's pcp should be updated.

Thanks
Wen Congyang

>
> Thanks
> Xishi Qiu
>
>>> >>
>>> >> vm_total_pages = nr_free_pagecache_pages();
>>> >> writeback_set_ratelimit();
>> >
>> >
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>