Hi Andrew and Andi,
We found a minor bug in the output of
/sys/devices/system/node/node[n]/meminfo
where the Active and Inactive values are in pages instead of Kbytes.
Looks like this occurred back in 2.6.20 when the code was changed
over to use node_page_state().
Thanks.
--- linux.2.6.26-rc8/drivers/base/node.c 2008-07-03 09:26:01.000000000 -0400
+++ new/drivers/base/node.c 2008-07-03 09:25:45.000000000 -0400
@@ -84,8 +84,8 @@ static ssize_t node_read_meminfo(struct
nid, K(i.totalram),
nid, K(i.freeram),
nid, K(i.totalram - i.freeram),
- nid, node_page_state(nid, NR_ACTIVE),
- nid, node_page_state(nid, NR_INACTIVE),
+ nid, K(node_page_state(nid, NR_ACTIVE)),
+ nid, K(node_page_state(nid, NR_INACTIVE)),
#ifdef CONFIG_HIGHMEM
nid, K(i.totalhigh),
nid, K(i.freehigh),
On Thu, 3 Jul 2008 10:35:57 -0400
John Blackwood <[email protected]> wrote:
> Hi Andrew and Andi,
>
> We found a minor bug in the output of
> /sys/devices/system/node/node[n]/meminfo
> where the Active and Inactive values are in pages instead of Kbytes.
>
> Looks like this occurred back in 2.6.20 when the code was changed
> over to use node_page_state().
>
> Thanks.
>
> --- linux.2.6.26-rc8/drivers/base/node.c 2008-07-03 09:26:01.000000000 -0400
> +++ new/drivers/base/node.c 2008-07-03 09:25:45.000000000 -0400
> @@ -84,8 +84,8 @@ static ssize_t node_read_meminfo(struct
> nid, K(i.totalram),
> nid, K(i.freeram),
> nid, K(i.totalram - i.freeram),
> - nid, node_page_state(nid, NR_ACTIVE),
> - nid, node_page_state(nid, NR_INACTIVE),
> + nid, K(node_page_state(nid, NR_ACTIVE)),
> + nid, K(node_page_state(nid, NR_INACTIVE)),
> #ifdef CONFIG_HIGHMEM
> nid, K(i.totalhigh),
> nid, K(i.freehigh),
ug. After all that time there might be applications out there which
use the existing numbers and work around the kernel bug. Which we'll
break if we fix it.
otoh, who the heck cares about the per-node active and inactive page
counts?
I guess we just fix this, backport the fix as far as we can and
apologise for any fallout. Not that there's likely to be any.
Please do remember to add a Signed-off-by: for kernel patches.
Andrew Morton wrote:
> On Thu, 3 Jul 2008 10:35:57 -0400
> John Blackwood <[email protected]> wrote:
>
>> Hi Andrew and Andi,
>>
>> We found a minor bug in the output of
>> /sys/devices/system/node/node[n]/meminfo
>> where the Active and Inactive values are in pages instead of Kbytes.
>>
>> Looks like this occurred back in 2.6.20 when the code was changed
>> over to use node_page_state().
>>
>> Thanks.
>>
>> --- linux.2.6.26-rc8/drivers/base/node.c 2008-07-03 09:26:01.000000000 -0400
>> +++ new/drivers/base/node.c 2008-07-03 09:25:45.000000000 -0400
>> @@ -84,8 +84,8 @@ static ssize_t node_read_meminfo(struct
>> nid, K(i.totalram),
>> nid, K(i.freeram),
>> nid, K(i.totalram - i.freeram),
>> - nid, node_page_state(nid, NR_ACTIVE),
>> - nid, node_page_state(nid, NR_INACTIVE),
>> + nid, K(node_page_state(nid, NR_ACTIVE)),
>> + nid, K(node_page_state(nid, NR_INACTIVE)),
>> #ifdef CONFIG_HIGHMEM
>> nid, K(i.totalhigh),
>> nid, K(i.freehigh),
>
> ug. After all that time there might be applications out there which
> use the existing numbers and work around the kernel bug. Which we'll
> break if we fix it.
true. our tool that looks at these numbers is going to be changed to
optionally multiply the values by 4 based on the kernel revision/version
number once this change makes it into the main tree.
> otoh, who the heck cares about the per-node active and inactive page
> counts?
:-)
> I guess we just fix this, backport the fix as far as we can and
> apologise for any fallout. Not that there's likely to be any.
>
o.k. Thanks.
> Please do remember to add a Signed-off-by: for kernel patches.
>
o.k. Sorry about that.