Hi,
I am running a 2.6.27 based embedded linux for MIPS. I can reproduce
a memory loss in kernel space by running a simple script
while true; do w >/dev/null; done
I am a VM newbie but it doesn't look like it's a leak. I notice
through /proc/zoneinfo that nr_free goes down while nr_active goes up
for DMA zone forever until the OOM killer starts killing tasks.
Here is a snapshot of /proc/zoneinfo at t0.
Node 0, zone DMA
pages free 708328
min 1919
low 2398
high 2878
scanned 0 (a: 0 i: 0)
spanned 1048064
present 935944
nr_free_pages 708328
nr_inactive 118115
nr_active 99415
nr_anon_pages 3891
nr_mapped 1747
nr_file_pages 125801
nr_dirty 0
nr_writeback 0
nr_slab_reclaimable 11158
nr_slab_unreclaimable 6594
nr_page_table_pages 362
nr_unstable 0
nr_bounce 0
nr_vmscan_write 0
nr_writeback_temp 0
protection: (0, 56, 56)
Here is another snapshot at t0 + 5 mins running while true; do
w>/dev/null; done.
Node 0, zone DMA
pages free 644875
min 1919
low 2398
high 2878
scanned 0 (a: 0 i: 0)
spanned 1048064
present 935944
nr_free_pages 644875
nr_inactive 118114
nr_active 163202
nr_anon_pages 4378
nr_mapped 1744
nr_file_pages 125805
nr_dirty 0
nr_writeback 0
nr_slab_reclaimable 11158
nr_slab_unreclaimable 6822
nr_page_table_pages 470
nr_unstable 0
nr_bounce 0
nr_vmscan_write 0
nr_writeback_temp 0
protection: (0, 56, 56)
It just feels like active pages are not being balanced correctly. Any
pointers troubleshooting this are highly appreciated.
Many thanks.
-Mohan
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>