2.5.66: task_struct memory leak?

2.5.66: task_struct memory leak?

[Russell King]

Hi,

With 2.5.66, I'm seeing what can only be described as a severe memory
leak.  This isn't something that I noticed on 2.5.65 - in fact, I have
several ARM machines happily running 2.5.65.

The leak seems to be centred around the task_struct slab, which seems
to do nothing but continually grow:

bash-2.04# grep task_struct /proc/slabinfo
task_struct         1868   1868    920  467  467    1 :   32   16 :   1868    1916   467    0    0    0   36 :   1509    496    139      0
...
bash-2.04# ps aux | wc -l
     20
bash-2.04# grep task_struct /proc/slabinfo
task_struct         1892   1892    920  473  473    1 :   32   16 :   1892    1956   473    0    0    0   36 :   1519    511    139      0
bash-2.04# grep task_struct /proc/slabinfo 
task_struct         1892   1892    920  473  473    1 :   32   16 :   1892    1957   473    0    0    0   36 :   1519    512    139      0
bash-2.04# grep task_struct /proc/slabinfo 
task_struct         1896   1896    920  474  474    1 :   32   16 :   1896    1961   474    0    0    0   36 :   1519    513    139      0
bash-2.04# grep task_struct /proc/slabinfo 
task_struct         1896   1896    920  474  474    1 :   32   16 :   1896    1961   474    0    0    0   36 :   1520    513    139      0
bash-2.04# grep task_struct /proc/slabinfo 
task_struct         1896   1896    920  474  474    1 :   32   16 :   1896    1961   474    0    0    0   36 :   1521    513    139      0
bash-2.04# grep task_struct /proc/slabinfo 
task_struct         1896   1896    920  474  474    1 :   32   16 :   1896    1961   474    0    0    0   36 :   1522    513    139      0
bash-2.04# grep task_struct /proc/slabinfo 
task_struct         1900   1900    920  475  475    1 :   32   16 :   1900    1965   475    0    0    0   36 :   1522    514    139      0

mm_struct seems to be fairly constant, so these are at least getting freed:
mm_struct             24     36    320    3    3    1 :   32   16 :    120     739    11    2    0    0   44 :   3144     53   3183      5

I'm seeing memory disappear at a rate of 8K / process, which seems to
suggest that the ARM level 1 page tables aren't getting freed either.

Is anyone seeing this type of behaviour on x86?

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

Re: 2.5.66: task_struct memory leak?

[Sean Neakums]

commence  Russell King quotation:

> I'm seeing memory disappear at a rate of 8K / process, which seems to
> suggest that the ARM level 1 page tables aren't getting freed either.
>
> Is anyone seeing this type of behaviour on x86?

Running 2.5.66 (plus Ingo Molnar's sched-2.5.66-A2), and I don't seem
to see a growth in the task_struct slab after repeated forking:

$ grep task_struct /proc/slabinfo ; for i in $(seq 1000) ; do /bin/true ; done ; grep task_struct /proc/slabinfo
task_struct           79     80   1552   16   16    2 :   54   27
task_struct           79     80   1552   16   16    2 :   54   27

-- 
Sean Neakums - <sneakums@zork.net>