diff options
author | Iliyan Malchev <malchev@google.com> | 2015-01-05 18:03:10 -0800 |
---|---|---|
committer | Iliyan Malchev <malchev@google.com> | 2015-01-05 18:03:10 -0800 |
commit | e4905c60dcc5fd45292573327bbe6baae54a8c2b (patch) | |
tree | d938972b033425494a490657e72a034ea3672ed6 /zImage-dtb | |
parent | 1b915d9591685b2b66fc6be8f542877202cc4c37 (diff) | |
download | hammerhead-kernel-e4905c60dcc5fd45292573327bbe6baae54a8c2b.tar.gz |
hammerhead: update prebuilt kernel
8399c4e ion_iommu_heap: allocate only with order=0
444f1ca msm: qpnp-power-on: print the power off reason
3521e06 sched,cgroup: Fix up task_groups list
The top commit eliminates the biggest contribution of highmem allocations of
order>=3. The stats below were collected with and without the patch, over
30-60 minutes, of calls to __alloc_pages_nodemask(), noting the count (in the
first column) of allocations with various GFP masks (in the second column as a
numbe and also pretty-printed in the fourth column), and breaking down total
count by order (in the third column). For example, the following line:
13 0x5200 1:1 2:1 3:11 NOWARN NORETRY COMP
means that there were 13 allocations with GFP mask 0x5200 (NOWARN NORETRY
COMP), of which 1 was order-1, 1 was order-2, and 11 were order-3.
Data with this patch:
up time: 01:01:10, idle time: 01:08:35, sleep time: 00:00:00
1 0x4020 2:1 HIGH COMP
1 0x52e0 3:1 HIGH IO FS NOWARN NORETRY COMP
1 0x85200 1:1 NOWARN NORETRY COMP RECLAIMABLE
1 0x85210 1:1 WAIT NOWARN NORETRY COMP RECLAIMABLE
1 0x85220 1:1 HIGH NOWARN NORETRY COMP RECLAIMABLE
2 0x252d0 2:2 WAIT IO FS NOWARN NORETRY COMP HARDWALL
2 0x82d0 4:2 WAIT IO FS NOWARN ZERO
4 0x80d0 4:1 6:1 1:1 3:1 WAIT IO FS ZERO
5 0x20 5:1 7:1 8:1 2:2 HIGH
12 0x15280 2:6 3:6 FS NOWARN NORETRY COMP NOMEMALLOC
13 0x5200 1:1 2:1 3:11 NOWARN NORETRY COMP
18 0x5220 1:13 2:2 3:3 HIGH NOWARN NORETRY COMP
108 0xc0d0 4:1 5:4 7:1 2:100 3:2 WAIT IO FS COMP ZERO
141 0x56d0 1:125 3:16 WAIT IO FS NOWARN REPEAT NORETRY COMP
448 0x85250 1:23 2:425 WAIT IO NOWARN NORETRY COMP RECLAIMABLE
506 0x405200 2:506 NOWARN NORETRY COMP NO_KSWAPD
1238 0x852d0 1:1238 WAIT IO FS NOWARN NORETRY COMP RECLAIMABLE
1812 0x52d0 1:690 2:108 3:1014 WAIT IO FS NOWARN NORETRY COMP
2068 0x40d0 4:7 5:7 6:2 2:2046 3:6 WAIT IO FS COMP
2521 0x44d0 2:2521 WAIT IO FS REPEAT COMP
33087 0xd0 1:31909 2:1178 WAIT IO FS
Data without this patch:
up time: 00:37:36, idle time: 00:41:25, sleep time: 00:00:00
1 0x252d0 2:1 WAIT IO FS NOWARN NORETRY COMP HARDWALL
1 0x4020 2:1 HIGH COMP
1 0x85200 1:1 NOWARN NORETRY COMP RECLAIMABLE
1 0x85220 1:1 HIGH NOWARN NORETRY COMP RECLAIMABLE
2 0x82d0 4:2 WAIT IO FS NOWARN ZERO
2 0x85210 1:2 WAIT NOWARN NORETRY COMP RECLAIMABLE
4 0x80d0 4:1 6:1 1:1 3:1 WAIT IO FS ZERO
5 0x20 5:1 7:1 8:1 2:2 HIGH
6 0x15280 2:6 FS NOWARN NORETRY COMP NOMEMALLOC
11 0x5220 1:3 2:5 3:3 HIGH NOWARN NORETRY COMP
15 0x5200 1:1 2:1 3:13 NOWARN NORETRY COMP
125 0xc0d0 4:1 5:4 7:1 2:117 3:2 WAIT IO FS COMP ZERO
191 0x56d0 1:172 2:2 3:17 WAIT IO FS NOWARN REPEAT NORETRY COMP
448 0x85250 1:23 2:425 WAIT IO NOWARN NORETRY COMP RECLAIMABLE
636 0x405200 2:636 NOWARN NORETRY COMP NO_KSWAPD
766 0x4092c2 4:734 8:32 HIGHMEM IO FS NOWARN NORETRY ZERO NO_KSWAPD
1084 0x852d0 1:1084 WAIT IO FS NOWARN NORETRY COMP RECLAIMABLE
1293 0x52d0 1:752 2:110 3:431 WAIT IO FS NOWARN NORETRY COMP
1576 0x40d0 4:7 5:7 6:2 2:1554 3:6 WAIT IO FS COMP
3122 0x44d0 2:3122 WAIT IO FS REPEAT COMP
31569 0xd0 1:30298 2:1271 WAIT IO FS
The following line:
766 0x4092c2 4:734 8:32 HIGHMEM IO FS NOWARN NORETRY ZERO NO_KSWAPD
shows that the patch eliminated 766 highmem allocations of orders 4 and 8.
This is by far the highest contribution of order>=3 of both high- and lowmem.
On a device that had run for ~30 minutes, we have only 2 order-3 buckets of
highmem:
Node 0, zone Normal 428 921 675 125 159 68 5 1 0 0 0
Node 0, zone HighMem 213 86 13 2 0 0 0 0 0 0 0
Most highmem is movable and is allocated from one page at a time (as both of
these data sets show, there aren't any other HIGHMEM allocations here of
order>=1), but these allocations are not movable; thus, over time, and with
heavy usage of media playback, we risk fragmenting highmem.
b/18729519 Watchdog bite in early boot
b/18069309 N5 running L can't run as many apps simultaneously as running K
Change-Id: I063f28956d7299001d9f5458885d5ba7047d7f42
Signed-off-by: Iliyan Malchev <malchev@google.com>
Diffstat (limited to 'zImage-dtb')
-rw-r--r-- | zImage-dtb | bin | 8474496 -> 8474216 bytes |
1 files changed, 0 insertions, 0 deletions
Binary files differ |