When we add a new node, the edge of memory may be wrong.
e.g. system has 4 nodes, and node3 is movable, node3 mem:[24G-32G],
1. hotremove the node3,
2. then hotadd node3 with a part of memory, mem:[26G-30G],
3. call hotadd_new_pgdat()
free_area_init_node()
get_pfn_range_for_nid()
4. it will return wrong start_pfn and end_pfn, because we have not
update the memblock.
This patch also fix another bug, please see
http://marc.info/?l=linux-kernel&m=142961156129456&w=2
Signed-off-by: Xishi Qiu <[email protected]>
---
mm/memory_hotplug.c | 3 +++
mm/page_alloc.c | 8 ++++++++
2 files changed, 11 insertions(+)
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 26fbba7..11f26cc 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -1269,6 +1269,7 @@ int __ref add_memory(int nid, u64 start, u64 size)
/* create new memmap entry */
firmware_map_add_hotplug(start, start + size, "System RAM");
+ memblock_add_node(start, size, nid);
goto out;
@@ -2005,6 +2006,8 @@ void __ref remove_memory(int nid, u64 start, u64 size)
/* remove memmap entry */
firmware_map_remove(start, start + size, "System RAM");
+ memblock_free(start, size);
+ memblock_remove(start, size);
arch_remove_memory(start, size);
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 1b91d37..7ca12b9 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -5050,6 +5050,10 @@ static unsigned long __meminit zone_spanned_pages_in_node(int nid,
{
unsigned long zone_start_pfn, zone_end_pfn;
+ /* When hotadd a new node, the node should be empty */
+ if (!node_start_pfn && !node_end_pfn)
+ return 0;
+
/* Get the start and end of the zone */
zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
@@ -5113,6 +5117,10 @@ static unsigned long __meminit zone_absent_pages_in_node(int nid,
unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
unsigned long zone_start_pfn, zone_end_pfn;
+ /* When hotadd a new node, the node should be empty */
+ if (!node_start_pfn && !node_end_pfn)
+ return 0;
+
zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
--
2.0.0
After hotadd_new_pgdat()->free_area_init_node(), the pgdat and zone's spanned/present
are both 0, so remove reset_node_managed_pages() and reset_node_managed_pages().
Signed-off-by: Xishi Qiu <[email protected]>
---
mm/memory_hotplug.c | 25 -------------------------
1 file changed, 25 deletions(-)
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 11f26cc..997dfad 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -1065,16 +1065,6 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typ
}
#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
-static void reset_node_present_pages(pg_data_t *pgdat)
-{
- struct zone *z;
-
- for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
- z->present_pages = 0;
-
- pgdat->node_present_pages = 0;
-}
-
/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
{
@@ -1109,21 +1099,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
build_all_zonelists(pgdat, NULL);
mutex_unlock(&zonelists_mutex);
- /*
- * zone->managed_pages is set to an approximate value in
- * free_area_init_core(), which will cause
- * /sys/device/system/node/nodeX/meminfo has wrong data.
- * So reset it to 0 before any memory is onlined.
- */
- reset_node_managed_pages(pgdat);
-
- /*
- * When memory is hot-added, all the memory is in offline state. So
- * clear all zones' present_pages because they will be updated in
- * online_pages() and offline_pages().
- */
- reset_node_present_pages(pgdat);
-
return pgdat;
}
--
2.0.0
Hi Shi,
Sorry for the late reply. I hope it won't be too late.
NON-ACK by me, I think.
I noticed that your first has been merged. But it won't fix the problem
these code intended to fix.
After your patch 1, zone's spanned/present won't be set to 0 because:
free_area_init_node()
|--> get_pfn_range_for_nid(&start_pfn, &end_pfn)
|--> calculate_node_totalpages(pgdat, start_pfn, end_pfn, ...)
| --> zone_spanned_pages_in_node()
| --> if (!node_start_pfn && !node_end_pfn) return
0; -------- false, won't return 0
| --> zone_absent_pages_in_node()
| --> if (!node_start_pfn && !node_end_pfn) return
0; -------- false, won't return 0
This is caused by a little bug in your patch 1.
You should put memblock_add_node(start, size, nid) before
hotadd_new_pgdat()
because:
hotadd_new_pgdat()
| --> free_area_init_node()
| --> get_pfn_range_for_nid()
| --> find memory ranges in memblock.
| --> memblock_add_node(start, size, nid) ------------------- if you
add it here, it doesn't work.
The result will be like below if we hotadd node 5.
[ 2007.577000] Initmem setup node 5 [mem
0x0000000000000000-0xffffffffffffffff]
[ 2007.584000] On node 5 totalpages: 0
[ 2007.585000] Built 5 zonelists in Node order, mobility grouping on.
Total pages: 32588823
[ 2007.594000] Policy zone: Normal
[ 2007.598000] init_memory_mapping: [mem 0x60000000000-0x607ffffffff]
And also, if we merge this patch, /sys/devices/system/node/nodeX/meminfo
will break.
Since this patch is not merged, I think let's just drop it.
And about the little bug in your patch 1, since I'm in a hurry, I have
already send a patch to fix it.
Thanks. :)
On 08/11/2015 03:33 PM, Xishi Qiu wrote:
> After hotadd_new_pgdat()->free_area_init_node(), the pgdat and zone's spanned/present
> are both 0, so remove reset_node_managed_pages() and reset_node_managed_pages().
>
> Signed-off-by: Xishi Qiu <[email protected]>
> ---
> mm/memory_hotplug.c | 25 -------------------------
> 1 file changed, 25 deletions(-)
>
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index 11f26cc..997dfad 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -1065,16 +1065,6 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typ
> }
> #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
>
> -static void reset_node_present_pages(pg_data_t *pgdat)
> -{
> - struct zone *z;
> -
> - for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
> - z->present_pages = 0;
> -
> - pgdat->node_present_pages = 0;
> -}
> -
> /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
> static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
> {
> @@ -1109,21 +1099,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
> build_all_zonelists(pgdat, NULL);
> mutex_unlock(&zonelists_mutex);
>
> - /*
> - * zone->managed_pages is set to an approximate value in
> - * free_area_init_core(), which will cause
> - * /sys/device/system/node/nodeX/meminfo has wrong data.
> - * So reset it to 0 before any memory is onlined.
> - */
> - reset_node_managed_pages(pgdat);
> -
> - /*
> - * When memory is hot-added, all the memory is in offline state. So
> - * clear all zones' present_pages because they will be updated in
> - * online_pages() and offline_pages().
> - */
> - reset_node_present_pages(pgdat);
> -
> return pgdat;
> }
>
On 2015/8/23 18:28, Tang Chen wrote:
> Hi Shi,
>
> Sorry for the late reply. I hope it won't be too late.
>
> NON-ACK by me, I think.
>
> I noticed that your first has been merged. But it won't fix the problem
> these code intended to fix.
>
> After your patch 1, zone's spanned/present won't be set to 0 because:
>
> free_area_init_node()
> |--> get_pfn_range_for_nid(&start_pfn, &end_pfn)
> |--> calculate_node_totalpages(pgdat, start_pfn, end_pfn, ...)
> | --> zone_spanned_pages_in_node()
> | --> if (!node_start_pfn && !node_end_pfn) return 0; -------- false, won't return 0
> | --> zone_absent_pages_in_node()
> | --> if (!node_start_pfn && !node_end_pfn) return 0; -------- false, won't return 0
>
Hi Tang,
Thank you for your reply. When we add a new node, it is not included in the
memblock, I have no idea why zone_spanned_pages_in_node() and zone_absent_pages_in_node()
won't return 0. Do you add some debug code and print it?
> This is caused by a little bug in your patch 1.
>
> You should put memblock_add_node(start, size, nid) before hotadd_new_pgdat()
> because:
>
My patch is just add an empty node first, later __add_zone() will update the size.
But it is all right put here, and it can fix the print bug.
> hotadd_new_pgdat()
> | --> free_area_init_node()
> | --> get_pfn_range_for_nid()
> | --> find memory ranges in memblock.
>
> | --> memblock_add_node(start, size, nid) ------------------- if you add it here, it doesn't work.
>
> The result will be like below if we hotadd node 5.
> [ 2007.577000] Initmem setup node 5 [mem 0x0000000000000000-0xffffffffffffffff]
pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
(u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
start_pfn and end_pfn are both 0, and 0-1 -> 0xffffffffffffffff, right?
> [ 2007.584000] On node 5 totalpages: 0
> [ 2007.585000] Built 5 zonelists in Node order, mobility grouping on. Total pages: 32588823
> [ 2007.594000] Policy zone: Normal
> [ 2007.598000] init_memory_mapping: [mem 0x60000000000-0x607ffffffff]
>
>
> And also, if we merge this patch, /sys/devices/system/node/nodeX/meminfo will break.
>
trigger call trace?
>
> Since this patch is not merged, I think let's just drop it.
>
> And about the little bug in your patch 1, since I'm in a hurry, I have already send a patch to fix it.
>
>
> Thanks. :)
>
>
> On 08/11/2015 03:33 PM, Xishi Qiu wrote:
>> After hotadd_new_pgdat()->free_area_init_node(), the pgdat and zone's spanned/present
>> are both 0, so remove reset_node_managed_pages() and reset_node_managed_pages().
>>
>> Signed-off-by: Xishi Qiu <[email protected]>
>> ---
>> mm/memory_hotplug.c | 25 -------------------------
>> 1 file changed, 25 deletions(-)
>>
>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>> index 11f26cc..997dfad 100644
>> --- a/mm/memory_hotplug.c
>> +++ b/mm/memory_hotplug.c
>> @@ -1065,16 +1065,6 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typ
>> }
>> #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
>> -static void reset_node_present_pages(pg_data_t *pgdat)
>> -{
>> - struct zone *z;
>> -
>> - for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
>> - z->present_pages = 0;
>> -
>> - pgdat->node_present_pages = 0;
>> -}
>> -
>> /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
>> static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
>> {
>> @@ -1109,21 +1099,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
>> build_all_zonelists(pgdat, NULL);
>> mutex_unlock(&zonelists_mutex);
>> - /*
>> - * zone->managed_pages is set to an approximate value in
>> - * free_area_init_core(), which will cause
>> - * /sys/device/system/node/nodeX/meminfo has wrong data.
>> - * So reset it to 0 before any memory is onlined.
>> - */
>> - reset_node_managed_pages(pgdat);
>> -
>> - /*
>> - * When memory is hot-added, all the memory is in offline state. So
>> - * clear all zones' present_pages because they will be updated in
>> - * online_pages() and offline_pages().
>> - */
>> - reset_node_present_pages(pgdat);
>> -
>> return pgdat;
>> }
>>
>
>
> .
>
On 08/24/2015 05:17 PM, Xishi Qiu wrote:
> On 2015/8/23 18:28, Tang Chen wrote:
>
>> Hi Shi,
>>
>> Sorry for the late reply. I hope it won't be too late.
>>
>> NON-ACK by me, I think.
>>
>> I noticed that your first has been merged. But it won't fix the problem
>> these code intended to fix.
>>
>> After your patch 1, zone's spanned/present won't be set to 0 because:
>>
>> free_area_init_node()
>> |--> get_pfn_range_for_nid(&start_pfn, &end_pfn)
>> |--> calculate_node_totalpages(pgdat, start_pfn, end_pfn, ...)
>> | --> zone_spanned_pages_in_node()
>> | --> if (!node_start_pfn && !node_end_pfn) return 0; -------- false, won't return 0
>> | --> zone_absent_pages_in_node()
>> | --> if (!node_start_pfn && !node_end_pfn) return 0; -------- false, won't return 0
>>
> Hi Tang,
>
> Thank you for your reply. When we add a new node, it is not included in the
> memblock, I have no idea why zone_spanned_pages_in_node() and zone_absent_pages_in_node()
> won't return 0. Do you add some debug code and print it?
Sorry, I didn't say it clearly since I was in a hurry.
If we move memblock_add_node() forward before hotadd_new_pgdat(),
then node_start_pfn and node_end_pfn won't be 0. And
zone_spanned_pages_in_node() and zone_absent_pages_in_node() won't
return 0.
In your patch, they will return 0. And that leads to the print problem.
>
>> This is caused by a little bug in your patch 1.
>>
>> You should put memblock_add_node(start, size, nid) before hotadd_new_pgdat()
>> because:
>>
> My patch is just add an empty node first, later __add_zone() will update the size.
> But it is all right put here, and it can fix the print bug.
Yes. I just found this bug in the log.
>
>> hotadd_new_pgdat()
>> | --> free_area_init_node()
>> | --> get_pfn_range_for_nid()
>> | --> find memory ranges in memblock.
>>
>> | --> memblock_add_node(start, size, nid) ------------------- if you add it here, it doesn't work.
>>
>> The result will be like below if we hotadd node 5.
>> [ 2007.577000] Initmem setup node 5 [mem 0x0000000000000000-0xffffffffffffffff]
> pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
> (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
> start_pfn and end_pfn are both 0, and 0-1 -> 0xffffffffffffffff, right?
Yes.
>
>> [ 2007.584000] On node 5 totalpages: 0
>> [ 2007.585000] Built 5 zonelists in Node order, mobility grouping on. Total pages: 32588823
>> [ 2007.594000] Policy zone: Normal
>> [ 2007.598000] init_memory_mapping: [mem 0x60000000000-0x607ffffffff]
>>
>>
>> And also, if we merge this patch, /sys/devices/system/node/nodeX/meminfo will break.
>>
> trigger call trace?
No. There is no error output. But if you see
/sys/devices/system/node/nodeX/meminfo,
memory size will double because totalpages is calculated once here, and
one more time
when onlining memory.
I think reset it to 0 in add_memory() is just a temporary solution. Will
improve it in the
future.
Thanks.
>
>> Since this patch is not merged, I think let's just drop it.
>>
>> And about the little bug in your patch 1, since I'm in a hurry, I have already send a patch to fix it.
>>
>>
>> Thanks. :)
>>
>>
>> On 08/11/2015 03:33 PM, Xishi Qiu wrote:
>>> After hotadd_new_pgdat()->free_area_init_node(), the pgdat and zone's spanned/present
>>> are both 0, so remove reset_node_managed_pages() and reset_node_managed_pages().
>>>
>>> Signed-off-by: Xishi Qiu <[email protected]>
>>> ---
>>> mm/memory_hotplug.c | 25 -------------------------
>>> 1 file changed, 25 deletions(-)
>>>
>>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>>> index 11f26cc..997dfad 100644
>>> --- a/mm/memory_hotplug.c
>>> +++ b/mm/memory_hotplug.c
>>> @@ -1065,16 +1065,6 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typ
>>> }
>>> #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
>>> -static void reset_node_present_pages(pg_data_t *pgdat)
>>> -{
>>> - struct zone *z;
>>> -
>>> - for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
>>> - z->present_pages = 0;
>>> -
>>> - pgdat->node_present_pages = 0;
>>> -}
>>> -
>>> /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
>>> static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
>>> {
>>> @@ -1109,21 +1099,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
>>> build_all_zonelists(pgdat, NULL);
>>> mutex_unlock(&zonelists_mutex);
>>> - /*
>>> - * zone->managed_pages is set to an approximate value in
>>> - * free_area_init_core(), which will cause
>>> - * /sys/device/system/node/nodeX/meminfo has wrong data.
>>> - * So reset it to 0 before any memory is onlined.
>>> - */
>>> - reset_node_managed_pages(pgdat);
>>> -
>>> - /*
>>> - * When memory is hot-added, all the memory is in offline state. So
>>> - * clear all zones' present_pages because they will be updated in
>>> - * online_pages() and offline_pages().
>>> - */
>>> - reset_node_present_pages(pgdat);
>>> -
>>> return pgdat;
>>> }
>>>
>>
>> .
>>
>
>
> .
>
On 2015/8/24 17:39, Tang Chen wrote:
>
> On 08/24/2015 05:17 PM, Xishi Qiu wrote:
>> On 2015/8/23 18:28, Tang Chen wrote:
>>
>>> Hi Shi,
>>>
>>> Sorry for the late reply. I hope it won't be too late.
>>>
>>> NON-ACK by me, I think.
>>>
>>> I noticed that your first has been merged. But it won't fix the problem
>>> these code intended to fix.
>>>
>>> After your patch 1, zone's spanned/present won't be set to 0 because:
>>>
>>> free_area_init_node()
>>> |--> get_pfn_range_for_nid(&start_pfn, &end_pfn)
>>> |--> calculate_node_totalpages(pgdat, start_pfn, end_pfn, ...)
>>> | --> zone_spanned_pages_in_node()
>>> | --> if (!node_start_pfn && !node_end_pfn) return 0; -------- false, won't return 0
>>> | --> zone_absent_pages_in_node()
>>> | --> if (!node_start_pfn && !node_end_pfn) return 0; -------- false, won't return 0
>>>
>> Hi Tang,
>>
>> Thank you for your reply. When we add a new node, it is not included in the
>> memblock, I have no idea why zone_spanned_pages_in_node() and zone_absent_pages_in_node()
>> won't return 0. Do you add some debug code and print it?
>
> Sorry, I didn't say it clearly since I was in a hurry.
>
> If we move memblock_add_node() forward before hotadd_new_pgdat(),
> then node_start_pfn and node_end_pfn won't be 0. And
> zone_spanned_pages_in_node() and zone_absent_pages_in_node() won't
> return 0.
>
> In your patch, they will return 0. And that leads to the print problem.
>
>>
>>> This is caused by a little bug in your patch 1.
>>>
>>> You should put memblock_add_node(start, size, nid) before hotadd_new_pgdat()
>>> because:
>>>
>> My patch is just add an empty node first, later __add_zone() will update the size.
>> But it is all right put here, and it can fix the print bug.
>
> Yes. I just found this bug in the log.
>
>>
>>> hotadd_new_pgdat()
>>> | --> free_area_init_node()
>>> | --> get_pfn_range_for_nid()
>>> | --> find memory ranges in memblock.
>>>
>>> | --> memblock_add_node(start, size, nid) ------------------- if you add it here, it doesn't work.
>>>
>>> The result will be like below if we hotadd node 5.
>>> [ 2007.577000] Initmem setup node 5 [mem 0x0000000000000000-0xffffffffffffffff]
>> pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
>> (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
>> start_pfn and end_pfn are both 0, and 0-1 -> 0xffffffffffffffff, right?
>
> Yes.
>
>>
>>> [ 2007.584000] On node 5 totalpages: 0
>>> [ 2007.585000] Built 5 zonelists in Node order, mobility grouping on. Total pages: 32588823
>>> [ 2007.594000] Policy zone: Normal
>>> [ 2007.598000] init_memory_mapping: [mem 0x60000000000-0x607ffffffff]
>>>
>>>
>>> And also, if we merge this patch, /sys/devices/system/node/nodeX/meminfo will break.
>>>
>> trigger call trace?
>
> No. There is no error output. But if you see /sys/devices/system/node/nodeX/meminfo,
> memory size will double because totalpages is calculated once here, and one more time
> when onlining memory.
>
Hi Tang,
Do you mean si_meminfo_node() -> val->totalram = managed_pages; will be added double?
But my patch will keep it 0 in hotadd_new_pgdat(), so it will not be double, right?
Thanks,
Xishi Qiu
> I think reset it to 0 in add_memory() is just a temporary solution. Will improve it in the
> future.
>
> Thanks.
>
>>
>>> Since this patch is not merged, I think let's just drop it.
>>>
>>> And about the little bug in your patch 1, since I'm in a hurry, I have already send a patch to fix it.
>>>
>>>
>>> Thanks. :)
>>>
>>>
>>> On 08/11/2015 03:33 PM, Xishi Qiu wrote:
>>>> After hotadd_new_pgdat()->free_area_init_node(), the pgdat and zone's spanned/present
>>>> are both 0, so remove reset_node_managed_pages() and reset_node_managed_pages().
>>>>
>>>> Signed-off-by: Xishi Qiu <[email protected]>
>>>> ---
>>>> mm/memory_hotplug.c | 25 -------------------------
>>>> 1 file changed, 25 deletions(-)
>>>>
>>>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>>>> index 11f26cc..997dfad 100644
>>>> --- a/mm/memory_hotplug.c
>>>> +++ b/mm/memory_hotplug.c
>>>> @@ -1065,16 +1065,6 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typ
>>>> }
>>>> #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
>>>> -static void reset_node_present_pages(pg_data_t *pgdat)
>>>> -{
>>>> - struct zone *z;
>>>> -
>>>> - for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
>>>> - z->present_pages = 0;
>>>> -
>>>> - pgdat->node_present_pages = 0;
>>>> -}
>>>> -
>>>> /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
>>>> static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
>>>> {
>>>> @@ -1109,21 +1099,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
>>>> build_all_zonelists(pgdat, NULL);
>>>> mutex_unlock(&zonelists_mutex);
>>>> - /*
>>>> - * zone->managed_pages is set to an approximate value in
>>>> - * free_area_init_core(), which will cause
>>>> - * /sys/device/system/node/nodeX/meminfo has wrong data.
>>>> - * So reset it to 0 before any memory is onlined.
>>>> - */
>>>> - reset_node_managed_pages(pgdat);
>>>> -
>>>> - /*
>>>> - * When memory is hot-added, all the memory is in offline state. So
>>>> - * clear all zones' present_pages because they will be updated in
>>>> - * online_pages() and offline_pages().
>>>> - */
>>>> - reset_node_present_pages(pgdat);
>>>> -
>>>> return pgdat;
>>>> }
>>>>
>>>
>>> .
>>>
>>
>>
>> .
>>
>
>
> .
>
On 08/24/2015 07:24 PM, Xishi Qiu wrote:
> ......
>>>> [ 2007.584000] On node 5 totalpages: 0
>>>> [ 2007.585000] Built 5 zonelists in Node order, mobility grouping on. Total pages: 32588823
>>>> [ 2007.594000] Policy zone: Normal
>>>> [ 2007.598000] init_memory_mapping: [mem 0x60000000000-0x607ffffffff]
>>>>
>>>>
>>>> And also, if we merge this patch, /sys/devices/system/node/nodeX/meminfo will break.
>>>>
>>> trigger call trace?
>> No. There is no error output. But if you see /sys/devices/system/node/nodeX/meminfo,
>> memory size will double because totalpages is calculated once here, and one more time
>> when onlining memory.
>>
> Hi Tang,
>
> Do you mean si_meminfo_node() -> val->totalram = managed_pages; will be added double?
> But my patch will keep it 0 in hotadd_new_pgdat(), so it will not be double, right?
>
Hi,
I mean this:
online_pages()
|--> zone->zone_pgdat->node_present_pages += onlined_pages;
It will be double.
Since meminfo data is retrieved from these kernel structures, /proc/meminfo
will be broken.
Actually speaking, reset it when hot-adding memory is not a good idea.
We should make the memory init code be suitable for both boot code and
memory hot-plug code.
Thanks.