GitHub/moto-9609/android_kernel_motorola_exynos9610.git
11 years agomemcg, vmscan: do not attempt soft limit reclaim if it would not scan anything
Michal Hocko [Thu, 12 Sep 2013 22:13:30 +0000 (15:13 -0700)]
memcg, vmscan: do not attempt soft limit reclaim if it would not scan anything

mem_cgroup_should_soft_reclaim controls whether soft reclaim pass is
done and it always says yes currently.  Memcg iterators are clever to
skip nodes that are not soft reclaimable quite efficiently but
mem_cgroup_should_soft_reclaim can be more clever and do not start the
soft reclaim pass at all if it knows that nothing would be scanned
anyway.

In order to do that, simply reuse mem_cgroup_soft_reclaim_eligible for
the target group of the reclaim and allow the pass only if the whole
subtree wouldn't be skipped.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Glauber Costa <glommer@openvz.org>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: track children in soft limit excess to improve soft limit
Michal Hocko [Thu, 12 Sep 2013 22:13:28 +0000 (15:13 -0700)]
memcg: track children in soft limit excess to improve soft limit

Soft limit reclaim has to check the whole reclaim hierarchy while doing
the first pass of the reclaim.  This leads to a higher system time which
can be visible especially when there are many groups in the hierarchy.

This patch adds a per-memcg counter of children in excess.  It also
restores MEM_CGROUP_TARGET_SOFTLIMIT into mem_cgroup_event_ratelimit for a
proper batching.

If a group crosses soft limit for the first time it increases parent's
children_in_excess up the hierarchy.  The similarly if a group gets below
the limit it will decrease the counter.  The transition phase is recorded
in soft_contributed flag.

mem_cgroup_soft_reclaim_eligible then uses this information to better
decide whether to skip the node or the whole subtree.  The rule is simple.
 Skip the node with a children in excess or skip the whole subtree
otherwise.

This has been tested by a stream IO (dd if=/dev/zero of=file with
4*MemTotal size) which is quite sensitive to overhead during reclaim.  The
load is running in a group with soft limit set to 0 and without any limit.
 Apart from that there was a hierarchy with ~500, 2k and 8k groups (two
groups on each level) without any pages in them.  base denotes to the
kernel on which the whole series is based on, rework is the kernel before
this patch and reworkoptim is with this patch applied:

* Run with soft limit set to 0
Elapsed
0-0-limit/base: min: 88.21 max: 94.61 avg: 91.73 std: 2.65 runs: 3
0-0-limit/rework: min: 76.05 [86.2%] max: 79.08 [83.6%] avg: 77.84 [84.9%] std: 1.30 runs: 3
0-0-limit/reworkoptim: min: 77.98 [88.4%] max: 80.36 [84.9%] avg: 78.92 [86.0%] std: 1.03 runs: 3
System
0.5k-0-limit/base: min: 34.86 max: 36.42 avg: 35.89 std: 0.73 runs: 3
0.5k-0-limit/rework: min: 43.26 [124.1%] max: 48.95 [134.4%] avg: 46.09 [128.4%] std: 2.32 runs: 3
0.5k-0-limit/reworkoptim: min: 46.98 [134.8%] max: 50.98 [140.0%] avg: 48.49 [135.1%] std: 1.77 runs: 3
Elapsed
0.5k-0-limit/base: min: 88.50 max: 97.52 avg: 93.92 std: 3.90 runs: 3
0.5k-0-limit/rework: min: 75.92 [85.8%] max: 78.45 [80.4%] avg: 77.34 [82.3%] std: 1.06 runs: 3
0.5k-0-limit/reworkoptim: min: 75.79 [85.6%] max: 79.37 [81.4%] avg: 77.55 [82.6%] std: 1.46 runs: 3
System
2k-0-limit/base: min: 34.57 max: 37.65 avg: 36.34 std: 1.30 runs: 3
2k-0-limit/rework: min: 64.17 [185.6%] max: 68.20 [181.1%] avg: 66.21 [182.2%] std: 1.65 runs: 3
2k-0-limit/reworkoptim: min: 49.78 [144.0%] max: 52.99 [140.7%] avg: 51.00 [140.3%] std: 1.42 runs: 3
Elapsed
2k-0-limit/base: min: 92.61 max: 97.83 avg: 95.03 std: 2.15 runs: 3
2k-0-limit/rework: min: 78.33 [84.6%] max: 84.08 [85.9%] avg: 81.09 [85.3%] std: 2.35 runs: 3
2k-0-limit/reworkoptim: min: 75.72 [81.8%] max: 78.57 [80.3%] avg: 76.73 [80.7%] std: 1.30 runs: 3
System
8k-0-limit/base: min: 39.78 max: 42.09 avg: 41.09 std: 0.97 runs: 3
8k-0-limit/rework: min: 200.86 [504.9%] max: 265.42 [630.6%] avg: 241.80 [588.5%] std: 29.06 runs: 3
8k-0-limit/reworkoptim: min: 53.70 [135.0%] max: 54.89 [130.4%] avg: 54.43 [132.5%] std: 0.52 runs: 3
Elapsed
8k-0-limit/base: min: 95.11 max: 98.61 avg: 96.81 std: 1.43 runs: 3
8k-0-limit/rework: min: 246.96 [259.7%] max: 331.47 [336.1%] avg: 301.32 [311.2%] std: 38.52 runs: 3
8k-0-limit/reworkoptim: min: 76.79 [80.7%] max: 81.71 [82.9%] avg: 78.97 [81.6%] std: 2.05 runs: 3

System time is increased by 30-40% but it is reduced a lot comparing to
kernel without this patch.  The higher time can be explained by the fact
that the original soft reclaim scanned at priority 0 so it was much more
effective for this workload (which is basically touch once and writeback).
 The Elapsed time looks better though (~20%).

* Run with no soft limit set
System
0-no-limit/base: min: 42.18 max: 50.38 avg: 46.44 std: 3.36 runs: 3
0-no-limit/rework: min: 40.57 [96.2%] max: 47.04 [93.4%] avg: 43.82 [94.4%] std: 2.64 runs: 3
0-no-limit/reworkoptim: min: 40.45 [95.9%] max: 45.28 [89.9%] avg: 42.10 [90.7%] std: 2.25 runs: 3
Elapsed
0-no-limit/base: min: 75.97 max: 78.21 avg: 76.87 std: 0.96 runs: 3
0-no-limit/rework: min: 75.59 [99.5%] max: 80.73 [103.2%] avg: 77.64 [101.0%] std: 2.23 runs: 3
0-no-limit/reworkoptim: min: 77.85 [102.5%] max: 82.42 [105.4%] avg: 79.64 [103.6%] std: 1.99 runs: 3
System
0.5k-no-limit/base: min: 44.54 max: 46.93 avg: 46.12 std: 1.12 runs: 3
0.5k-no-limit/rework: min: 42.09 [94.5%] max: 46.16 [98.4%] avg: 43.92 [95.2%] std: 1.69 runs: 3
0.5k-no-limit/reworkoptim: min: 42.47 [95.4%] max: 45.67 [97.3%] avg: 44.06 [95.5%] std: 1.31 runs: 3
Elapsed
0.5k-no-limit/base: min: 78.26 max: 81.49 avg: 79.65 std: 1.36 runs: 3
0.5k-no-limit/rework: min: 77.01 [98.4%] max: 80.43 [98.7%] avg: 78.30 [98.3%] std: 1.52 runs: 3
0.5k-no-limit/reworkoptim: min: 76.13 [97.3%] max: 77.87 [95.6%] avg: 77.18 [96.9%] std: 0.75 runs: 3
System
2k-no-limit/base: min: 62.96 max: 69.14 avg: 66.14 std: 2.53 runs: 3
2k-no-limit/rework: min: 76.01 [120.7%] max: 81.06 [117.2%] avg: 78.17 [118.2%] std: 2.12 runs: 3
2k-no-limit/reworkoptim: min: 62.57 [99.4%] max: 66.10 [95.6%] avg: 64.53 [97.6%] std: 1.47 runs: 3
Elapsed
2k-no-limit/base: min: 76.47 max: 84.22 avg: 79.12 std: 3.60 runs: 3
2k-no-limit/rework: min: 89.67 [117.3%] max: 93.26 [110.7%] avg: 91.10 [115.1%] std: 1.55 runs: 3
2k-no-limit/reworkoptim: min: 76.94 [100.6%] max: 79.21 [94.1%] avg: 78.45 [99.2%] std: 1.07 runs: 3
System
8k-no-limit/base: min: 104.74 max: 151.34 avg: 129.21 std: 19.10 runs: 3
8k-no-limit/rework: min: 205.23 [195.9%] max: 285.94 [188.9%] avg: 258.98 [200.4%] std: 38.01 runs: 3
8k-no-limit/reworkoptim: min: 161.16 [153.9%] max: 184.54 [121.9%] avg: 174.52 [135.1%] std: 9.83 runs: 3
Elapsed
8k-no-limit/base: min: 125.43 max: 181.00 avg: 154.81 std: 22.80 runs: 3
8k-no-limit/rework: min: 254.05 [202.5%] max: 355.67 [196.5%] avg: 321.46 [207.6%] std: 47.67 runs: 3
8k-no-limit/reworkoptim: min: 193.77 [154.5%] max: 222.72 [123.0%] avg: 210.18 [135.8%] std: 12.13 runs: 3

Both System and Elapsed are in stdev with the base kernel for all
configurations except for 8k where both System and Elapsed are up by 35%.
I do not have a good explanation for this because there is no soft reclaim
pass going on as no group is above the limit which is checked in
mem_cgroup_should_soft_reclaim.

Then I have tested kernel build with the same configuration to see the
behavior with a more general behavior.

* Soft limit set to 0 for the build
System
0-0-limit/base: min: 242.70 max: 245.17 avg: 243.85 std: 1.02 runs: 3
0-0-limit/rework min: 237.86 [98.0%] max: 240.22 [98.0%] avg: 239.00 [98.0%] std: 0.97 runs: 3
0-0-limit/reworkoptim: min: 241.11 [99.3%] max: 243.53 [99.3%] avg: 242.01 [99.2%] std: 1.08 runs: 3
Elapsed
0-0-limit/base: min: 348.48 max: 360.86 avg: 356.04 std: 5.41 runs: 3
0-0-limit/rework min: 286.95 [82.3%] max: 290.26 [80.4%] avg: 288.27 [81.0%] std: 1.43 runs: 3
0-0-limit/reworkoptim: min: 286.55 [82.2%] max: 289.00 [80.1%] avg: 287.69 [80.8%] std: 1.01 runs: 3
System
0.5k-0-limit/base: min: 251.77 max: 254.41 avg: 252.70 std: 1.21 runs: 3
0.5k-0-limit/rework min: 286.44 [113.8%] max: 289.30 [113.7%] avg: 287.60 [113.8%] std: 1.23 runs: 3
0.5k-0-limit/reworkoptim: min: 252.18 [100.2%] max: 253.16 [99.5%] avg: 252.62 [100.0%] std: 0.41 runs: 3
Elapsed
0.5k-0-limit/base: min: 347.83 max: 353.06 avg: 350.04 std: 2.21 runs: 3
0.5k-0-limit/rework min: 290.19 [83.4%] max: 295.62 [83.7%] avg: 293.12 [83.7%] std: 2.24 runs: 3
0.5k-0-limit/reworkoptim: min: 293.91 [84.5%] max: 294.87 [83.5%] avg: 294.29 [84.1%] std: 0.42 runs: 3
System
2k-0-limit/base: min: 263.05 max: 271.52 avg: 267.94 std: 3.58 runs: 3
2k-0-limit/rework min: 458.99 [174.5%] max: 468.31 [172.5%] avg: 464.45 [173.3%] std: 3.97 runs: 3
2k-0-limit/reworkoptim: min: 267.10 [101.5%] max: 279.38 [102.9%] avg: 272.78 [101.8%] std: 5.05 runs: 3
Elapsed
2k-0-limit/base: min: 372.33 max: 379.32 avg: 375.47 std: 2.90 runs: 3
2k-0-limit/rework min: 334.40 [89.8%] max: 339.52 [89.5%] avg: 337.44 [89.9%] std: 2.20 runs: 3
2k-0-limit/reworkoptim: min: 301.47 [81.0%] max: 319.19 [84.1%] avg: 307.90 [82.0%] std: 8.01 runs: 3
System
8k-0-limit/base: min: 320.50 max: 332.10 avg: 325.46 std: 4.88 runs: 3
8k-0-limit/rework min: 1115.76 [348.1%] max: 1165.66 [351.0%] avg: 1132.65 [348.0%] std: 23.34 runs: 3
8k-0-limit/reworkoptim: min: 403.75 [126.0%] max: 409.22 [123.2%] avg: 406.16 [124.8%] std: 2.28 runs: 3
Elapsed
8k-0-limit/base: min: 475.48 max: 585.19 avg: 525.54 std: 45.30 runs: 3
8k-0-limit/rework min: 616.25 [129.6%] max: 625.90 [107.0%] avg: 620.68 [118.1%] std: 3.98 runs: 3
8k-0-limit/reworkoptim: min: 420.18 [88.4%] max: 428.28 [73.2%] avg: 423.05 [80.5%] std: 3.71 runs: 3

Apart from 8k the system time is comparable with the base kernel while
Elapsed is up to 20% better with all configurations.

* No soft limit set
System
0-no-limit/base: min: 234.76 max: 237.42 avg: 236.25 std: 1.11 runs: 3
0-no-limit/rework min: 233.09 [99.3%] max: 238.65 [100.5%] avg: 236.09 [99.9%] std: 2.29 runs: 3
0-no-limit/reworkoptim: min: 236.12 [100.6%] max: 240.53 [101.3%] avg: 237.94 [100.7%] std: 1.88 runs: 3
Elapsed
0-no-limit/base: min: 288.52 max: 295.42 avg: 291.29 std: 2.98 runs: 3
0-no-limit/rework min: 283.17 [98.1%] max: 284.33 [96.2%] avg: 283.78 [97.4%] std: 0.48 runs: 3
0-no-limit/reworkoptim: min: 288.50 [100.0%] max: 290.79 [98.4%] avg: 289.78 [99.5%] std: 0.95 runs: 3
System
0.5k-no-limit/base: min: 286.51 max: 293.23 avg: 290.21 std: 2.78 runs: 3
0.5k-no-limit/rework min: 291.69 [101.8%] max: 294.38 [100.4%] avg: 292.97 [101.0%] std: 1.10 runs: 3
0.5k-no-limit/reworkoptim: min: 277.05 [96.7%] max: 288.76 [98.5%] avg: 284.17 [97.9%] std: 5.11 runs: 3
Elapsed
0.5k-no-limit/base: min: 294.94 max: 298.92 avg: 296.47 std: 1.75 runs: 3
0.5k-no-limit/rework min: 292.55 [99.2%] max: 294.21 [98.4%] avg: 293.55 [99.0%] std: 0.72 runs: 3
0.5k-no-limit/reworkoptim: min: 294.41 [99.8%] max: 301.67 [100.9%] avg: 297.78 [100.4%] std: 2.99 runs: 3
System
2k-no-limit/base: min: 443.41 max: 466.66 avg: 457.66 std: 10.19 runs: 3
2k-no-limit/rework min: 490.11 [110.5%] max: 516.02 [110.6%] avg: 501.42 [109.6%] std: 10.83 runs: 3
2k-no-limit/reworkoptim: min: 435.25 [98.2%] max: 458.11 [98.2%] avg: 446.73 [97.6%] std: 9.33 runs: 3
Elapsed
2k-no-limit/base: min: 330.85 max: 333.75 avg: 332.52 std: 1.23 runs: 3
2k-no-limit/rework min: 343.06 [103.7%] max: 349.59 [104.7%] avg: 345.95 [104.0%] std: 2.72 runs: 3
2k-no-limit/reworkoptim: min: 330.01 [99.7%] max: 333.92 [100.1%] avg: 332.22 [99.9%] std: 1.64 runs: 3
System
8k-no-limit/base: min: 1175.64 max: 1259.38 avg: 1222.39 std: 34.88 runs: 3
8k-no-limit/rework min: 1226.31 [104.3%] max: 1241.60 [98.6%] avg: 1233.74 [100.9%] std: 6.25 runs: 3
8k-no-limit/reworkoptim: min: 1023.45 [87.1%] max: 1056.74 [83.9%] avg: 1038.92 [85.0%] std: 13.69 runs: 3
Elapsed
8k-no-limit/base: min: 613.36 max: 619.60 avg: 616.47 std: 2.55 runs: 3
8k-no-limit/rework min: 627.56 [102.3%] max: 642.33 [103.7%] avg: 633.44 [102.8%] std: 6.39 runs: 3
8k-no-limit/reworkoptim: min: 545.89 [89.0%] max: 555.36 [89.6%] avg: 552.06 [89.6%] std: 4.37 runs: 3

and these numbers look good as well.  System time is around 100%
(suprisingly better for the 8k case) and Elapsed is copies that trend.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Glauber Costa <glommer@openvz.org>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: enhance memcg iterator to support predicates
Michal Hocko [Thu, 12 Sep 2013 22:13:26 +0000 (15:13 -0700)]
memcg: enhance memcg iterator to support predicates

The caller of the iterator might know that some nodes or even subtrees
should be skipped but there is no way to tell iterators about that so the
only choice left is to let iterators to visit each node and do the
selection outside of the iterating code.  This, however, doesn't scale
well with hierarchies with many groups where only few groups are
interesting.

This patch adds mem_cgroup_iter_cond variant of the iterator with a
callback which gets called for every visited node.  There are three
possible ways how the callback can influence the walk.  Either the node is
visited, it is skipped but the tree walk continues down the tree or the
whole subtree of the current group is skipped.

[hughd@google.com: fix memcg-less page reclaim]
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Glauber Costa <glommer@openvz.org>
Cc: Greg Thelen <gthelen@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agovmscan, memcg: do softlimit reclaim also for targeted reclaim
Michal Hocko [Thu, 12 Sep 2013 22:13:25 +0000 (15:13 -0700)]
vmscan, memcg: do softlimit reclaim also for targeted reclaim

Soft reclaim has been done only for the global reclaim (both background
and direct).  Since "memcg: integrate soft reclaim tighter with zone
shrinking code" there is no reason for this limitation anymore as the soft
limit reclaim doesn't use any special code paths and it is a part of the
zone shrinking code which is used by both global and targeted reclaims.

From the semantic point of view it is natural to consider soft limit
before touching all groups in the hierarchy tree which is touching the
hard limit because soft limit tells us where to push back when there is a
memory pressure.  It is not important whether the pressure comes from the
limit or imbalanced zones.

This patch simply enables soft reclaim unconditionally in
mem_cgroup_should_soft_reclaim so it is enabled for both global and
targeted reclaim paths.  mem_cgroup_soft_reclaim_eligible needs to learn
about the root of the reclaim to know where to stop checking soft limit
state of parents up the hierarchy.  Say we have

A (over soft limit)
 \
  B (below s.l., hit the hard limit)
 / \
C   D (below s.l.)

B is the source of the outside memory pressure now for D but we shouldn't
soft reclaim it because it is behaving well under B subtree and we can
still reclaim from C (pressumably it is over the limit).
mem_cgroup_soft_reclaim_eligible should therefore stop climbing up the
hierarchy at B (root of the memory pressure).

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Glauber Costa <glommer@openvz.org>
Reviewed-by: Tejun Heo <tj@kernel.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: get rid of soft-limit tree infrastructure
Michal Hocko [Thu, 12 Sep 2013 22:13:23 +0000 (15:13 -0700)]
memcg: get rid of soft-limit tree infrastructure

Now that the soft limit is integrated to the reclaim directly the whole
soft-limit tree infrastructure is not needed anymore.  Rip it out.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Glauber Costa <glommer@openvz.org>
Reviewed-by: Tejun Heo <tj@kernel.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg, vmscan: integrate soft reclaim tighter with zone shrinking code
Michal Hocko [Thu, 12 Sep 2013 22:13:21 +0000 (15:13 -0700)]
memcg, vmscan: integrate soft reclaim tighter with zone shrinking code

This patchset is sitting out of tree for quite some time without any
objections.  I would be really happy if it made it into 3.12.  I do not
want to push it too hard but I think this work is basically ready and
waiting more doesn't help.

The basic idea is quite simple.  Pull soft reclaim into shrink_zone in the
first step and get rid of the previous soft reclaim infrastructure.
shrink_zone is done in two passes now.  First it tries to do the soft
limit reclaim and it falls back to reclaim-all mode if no group is over
the limit or no pages have been scanned.  The second pass happens at the
same priority so the only time we waste is the memcg tree walk which has
been updated in the third step to have only negligible overhead.

As a bonus we will get rid of a _lot_ of code by this and soft reclaim
will not stand out like before when it wasn't integrated into the zone
shrinking code and it reclaimed at priority 0 (the testing results show
that some workloads suffers from such an aggressive reclaim).  The clean
up is in a separate patch because I felt it would be easier to review that
way.

The second step is soft limit reclaim integration into targeted reclaim.
It should be rather straight forward.  Soft limit has been used only for
the global reclaim so far but it makes sense for any kind of pressure
coming from up-the-hierarchy, including targeted reclaim.

The third step (patches 4-8) addresses the tree walk overhead by enhancing
memcg iterators to enable skipping whole subtrees and tracking number of
over soft limit children at each level of the hierarchy.  This information
is updated same way the old soft limit tree was updated (from
memcg_check_events) so we shouldn't see an additional overhead.  In fact
mem_cgroup_update_soft_limit is much simpler than tree manipulation done
previously.

__shrink_zone uses mem_cgroup_soft_reclaim_eligible as a predicate for
mem_cgroup_iter so the decision whether a particular group should be
visited is done at the iterator level which allows us to decide to skip
the whole subtree as well (if there is no child in excess).  This reduces
the tree walk overhead considerably.

* TEST 1
========

My primary test case was a parallel kernel build with 2 groups (make is
running with -j8 with a distribution .config in a separate cgroup without
any hard limit) on a 32 CPU machine booted with 1GB memory and both builds
run taskset to Node 0 cpus.

I was mostly interested in 2 setups.  Default - no soft limit set and -
and 0 soft limit set to both groups.  The first one should tell us whether
the rework regresses the default behavior while the second one should show
us improvements in an extreme case where both workloads are always over
the soft limit.

/usr/bin/time -v has been used to collect the statistics and each
configuration had 3 runs after fresh boot without any other load on the
system.

base is mmotm-2013-07-18-16-40
rework all 8 patches applied on top of base

* No-limit
User
no-limit/base: min: 651.92 max: 672.65 avg: 664.33 std: 8.01 runs: 6
no-limit/rework: min: 657.34 [100.8%] max: 668.39 [99.4%] avg: 663.13 [99.8%] std: 3.61 runs: 6
System
no-limit/base: min: 69.33 max: 71.39 avg: 70.32 std: 0.79 runs: 6
no-limit/rework: min: 69.12 [99.7%] max: 71.05 [99.5%] avg: 70.04 [99.6%] std: 0.59 runs: 6
Elapsed
no-limit/base: min: 398.27 max: 422.36 avg: 408.85 std: 7.74 runs: 6
no-limit/rework: min: 386.36 [97.0%] max: 438.40 [103.8%] avg: 416.34 [101.8%] std: 18.85 runs: 6

The results are within noise. Elapsed time has a bigger variance but the
average looks good.

* 0-limit
User
0-limit/base: min: 573.76 max: 605.63 avg: 585.73 std: 12.21 runs: 6
0-limit/rework: min: 645.77 [112.6%] max: 666.25 [110.0%] avg: 656.97 [112.2%] std: 7.77 runs: 6
System
0-limit/base: min: 69.57 max: 71.13 avg: 70.29 std: 0.54 runs: 6
0-limit/rework: min: 68.68 [98.7%] max: 71.40 [100.4%] avg: 69.91 [99.5%] std: 0.87 runs: 6
Elapsed
0-limit/base: min: 1306.14 max: 1550.17 avg: 1430.35 std: 90.86 runs: 6
0-limit/rework: min: 404.06 [30.9%] max: 465.94 [30.1%] avg: 434.81 [30.4%] std: 22.68 runs: 6

The improvement is really huge here (even bigger than with my previous
testing and I suspect that this highly depends on the storage).  Page
fault statistics tell us at least part of the story:

Minor
0-limit/base: min: 37180461.00 max: 37319986.00 avg: 37247470.00 std: 54772.71 runs: 6
0-limit/rework: min: 36751685.00 [98.8%] max: 36805379.00 [98.6%] avg: 36774506.33 [98.7%] std: 17109.03 runs: 6
Major
0-limit/base: min: 170604.00 max: 221141.00 avg: 196081.83 std: 18217.01 runs: 6
0-limit/rework: min: 2864.00 [1.7%] max: 10029.00 [4.5%] avg: 5627.33 [2.9%] std: 2252.71 runs: 6

Same as with my previous testing Minor faults are more or less within
noise but Major fault count is way bellow the base kernel.

While this looks as a nice win it is fair to say that 0-limit
configuration is quite artificial. So I was playing with 0-no-limit
loads as well.

* TEST 2
========

The following results are from 2 groups configuration on a 16GB machine
(single NUMA node).

- A running stream IO (dd if=/dev/zero of=local.file bs=1024) with
  2*TotalMem with 0 soft limit.
- B running a mem_eater which consumes TotalMem-1G without any limit. The
  mem_eater consumes the memory in 100 chunks with 1s nap after each
  mmap+poppulate so that both loads have chance to fight for the memory.

The expected result is that B shouldn't be reclaimed and A shouldn't see
a big dropdown in elapsed time.

User
base: min: 2.68 max: 2.89 avg: 2.76 std: 0.09 runs: 3
rework: min: 3.27 [122.0%] max: 3.74 [129.4%] avg: 3.44 [124.6%] std: 0.21 runs: 3
System
base: min: 86.26 max: 88.29 avg: 87.28 std: 0.83 runs: 3
rework: min: 81.05 [94.0%] max: 84.96 [96.2%] avg: 83.14 [95.3%] std: 1.61 runs: 3
Elapsed
base: min: 317.28 max: 332.39 avg: 325.84 std: 6.33 runs: 3
rework: min: 281.53 [88.7%] max: 298.16 [89.7%] avg: 290.99 [89.3%] std: 6.98 runs: 3

System time improved slightly as well as Elapsed. My previous testing
has shown worse numbers but this again seem to depend on the storage
speed.

My theory is that the writeback doesn't catch up and prio-0 soft reclaim
falls into wait on writeback page too often in the base kernel. The
patched kernel doesn't do that because the soft reclaim is done from the
kswapd/direct reclaim context. This can be seen on the following graph
nicely. The A's group usage_in_bytes regurarly drops really low very often.

All 3 runs
http://labs.suse.cz/mhocko/soft_limit_rework/stream_io-vs-mem_eater/stream.png
resp. a detail of the single run
http://labs.suse.cz/mhocko/soft_limit_rework/stream_io-vs-mem_eater/stream-one-run.png

mem_eater seems to be doing better as well. It gets to the full
allocation size faster as can be seen on the following graph:
http://labs.suse.cz/mhocko/soft_limit_rework/stream_io-vs-mem_eater/mem_eater-one-run.png

/proc/meminfo collected during the test also shows that rework kernel
hasn't swapped that much (well almost not at all):
base: max: 123900 K avg: 56388.29 K
rework: max: 300 K avg: 128.68 K

kswapd and direct reclaim statistics are of no use unfortunatelly because
soft reclaim is not accounted properly as the counters are hidden by
global_reclaim() checks in the base kernel.

* TEST 3
========

Another test was the same configuration as TEST2 except the stream IO was
replaced by a single kbuild (16 parallel jobs bound to Node0 cpus same as
in TEST1) and mem_eater allocated TotalMem-200M so kbuild had only 200MB
left.

Kbuild did better with the rework kernel here as well:
User
base: min: 860.28 max: 872.86 avg: 868.03 std: 5.54 runs: 3
rework: min: 880.81 [102.4%] max: 887.45 [101.7%] avg: 883.56 [101.8%] std: 2.83 runs: 3
System
base: min: 84.35 max: 85.06 avg: 84.79 std: 0.31 runs: 3
rework: min: 85.62 [101.5%] max: 86.09 [101.2%] avg: 85.79 [101.2%] std: 0.21 runs: 3
Elapsed
base: min: 135.36 max: 243.30 avg: 182.47 std: 45.12 runs: 3
rework: min: 110.46 [81.6%] max: 116.20 [47.8%] avg: 114.15 [62.6%] std: 2.61 runs: 3
Minor
base: min: 36635476.00 max: 36673365.00 avg: 36654812.00 std: 15478.03 runs: 3
rework: min: 36639301.00 [100.0%] max: 36695541.00 [100.1%] avg: 36665511.00 [100.0%] std: 23118.23 runs: 3
Major
base: min: 14708.00 max: 53328.00 avg: 31379.00 std: 16202.24 runs: 3
rework: min: 302.00 [2.1%] max: 414.00 [0.8%] avg: 366.33 [1.2%] std: 47.22 runs: 3

Again we can see a significant improvement in Elapsed (it also seems to
be more stable), there is a huge dropdown for the Major page faults and
much more swapping:
base: max: 583736 K avg: 112547.43 K
rework: max: 4012 K avg: 124.36 K

Graphs from all three runs show the variability of the kbuild quite
nicely.  It even seems that it took longer after every run with the base
kernel which would be quite surprising as the source tree for the build is
removed and caches are dropped after each run so the build operates on a
freshly extracted sources everytime.
http://labs.suse.cz/mhocko/soft_limit_rework/stream_io-vs-mem_eater/kbuild-mem_eater.png

My other testing shows that this is just a matter of timing and other runs
behave differently the std for Elapsed time is similar ~50.  Example of
other three runs:
http://labs.suse.cz/mhocko/soft_limit_rework/stream_io-vs-mem_eater/kbuild-mem_eater2.png

So to wrap this up.  The series is still doing good and improves the soft
limit.

The testing results for bunch of cgroups with both stream IO and kbuild
loads can be found in "memcg: track children in soft limit excess to
improve soft limit".

This patch:

Memcg soft reclaim has been traditionally triggered from the global
reclaim paths before calling shrink_zone.  mem_cgroup_soft_limit_reclaim
then picked up a group which exceeds the soft limit the most and reclaimed
it with 0 priority to reclaim at least SWAP_CLUSTER_MAX pages.

The infrastructure requires per-node-zone trees which hold over-limit
groups and keep them up-to-date (via memcg_check_events) which is not cost
free.  Although this overhead hasn't turned out to be a bottle neck the
implementation is suboptimal because mem_cgroup_update_tree has no idea
which zones consumed memory over the limit so we could easily end up
having a group on a node-zone tree having only few pages from that
node-zone.

This patch doesn't try to fix node-zone trees management because it seems
that integrating soft reclaim into zone shrinking sounds much easier and
more appropriate for several reasons.  First of all 0 priority reclaim was
a crude hack which might lead to big stalls if the group's LRUs are big
and hard to reclaim (e.g.  a lot of dirty/writeback pages).  Soft reclaim
should be applicable also to the targeted reclaim which is awkward right
now without additional hacks.  Last but not least the whole infrastructure
eats quite some code.

After this patch shrink_zone is done in 2 passes.  First it tries to do
the soft reclaim if appropriate (only for global reclaim for now to keep
compatible with the original state) and fall back to ignoring soft limit
if no group is eligible to soft reclaim or nothing has been scanned during
the first pass.  Only groups which are over their soft limit or any of
their parents up the hierarchy is over the limit are considered eligible
during the first pass.

Soft limit tree which is not necessary anymore will be removed in the
follow up patch to make this patch smaller and easier to review.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Glauber Costa <glommer@openvz.org>
Reviewed-by: Tejun Heo <tj@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ying Han <yinghan@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Glauber Costa <glommer@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemcg: remove redundant code in mem_cgroup_force_empty_write()
Li Zefan [Thu, 12 Sep 2013 22:13:19 +0000 (15:13 -0700)]
memcg: remove redundant code in mem_cgroup_force_empty_write()

vfs guarantees the cgroup won't be destroyed, so it's redundant to get a
css reference.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoBye, bye, WfW flag
Linus Torvalds [Thu, 12 Sep 2013 02:55:12 +0000 (19:55 -0700)]
Bye, bye, WfW flag

This reverts the Linux for Workgroups thing.  And no, before somebody
asks, we're not doing Linux95.  Not for a few years, at least.

Sure, the flag added some color to the logo, and could have remained as
a testament to my leet gimp skills.  But no.  And I'll do this early, to
avoid the chance of forgetting when I'm doing the actual rc1 release on
the road.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge tag 'ecryptfs-3.12-rc1-crypt-ctx' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 12 Sep 2013 02:17:04 +0000 (19:17 -0700)]
Merge tag 'ecryptfs-3.12-rc1-crypt-ctx' of git://git./linux/kernel/git/tyhicks/ecryptfs

Pull eCryptfs fixes from Tyler Hicks:
 "Two small fixes to the code that initializes the per-file crypto
  contexts"

* tag 'ecryptfs-3.12-rc1-crypt-ctx' of git://git.kernel.org/pub/scm/linux/kernel/git/tyhicks/ecryptfs:
  ecryptfs: avoid ctx initialization race
  ecryptfs: remove check for if an array is NULL

11 years agoMerge branch 'for-v3.12-fix' of git://git.linaro.org/people/mszyprowski/linux-dma...
Linus Torvalds [Thu, 12 Sep 2013 02:11:59 +0000 (19:11 -0700)]
Merge branch 'for-v3.12-fix' of git://git.linaro.org/people/mszyprowski/linux-dma-mapping

Pull DMA-mapping fix from Marek Szyprowski:
 "A build bugfix for the device tree support for reserved memory
  regions.  Due to superfluous include the common code failed to build
  on ARM64 and MIPS architectures.

  The patch that caused the build break has lived at linux-next for
  about two weeks and noone noticed the issue, what convinced me that
  everything was ok"

* 'for-v3.12-fix' of git://git.linaro.org/people/mszyprowski/linux-dma-mapping:
  drivers: of: fix build break if asm/dma-contiguous.h is missing

11 years agoMerge tag 'for-3.12' of git://git.linaro.org/people/sumitsemwal/linux-dma-buf
Linus Torvalds [Thu, 12 Sep 2013 02:09:14 +0000 (19:09 -0700)]
Merge tag 'for-3.12' of git://git.linaro.org/people/sumitsemwal/linux-dma-buf

Pull dma-buf updates from Sumit Semwal:
 "Yet another small one - dma-buf framework now supports size discovery
  of the buffer via llseek"

* tag 'for-3.12' of git://git.linaro.org/people/sumitsemwal/linux-dma-buf:
  dma-buf: Expose buffer size to userspace (v2)
  dma-buf: Check return value of anon_inode_getfile

11 years agoMerge branch 'akpm' (patches from Andrew Morton)
Linus Torvalds [Wed, 11 Sep 2013 23:08:54 +0000 (16:08 -0700)]
Merge branch 'akpm' (patches from Andrew Morton)

Merge first patch-bomb from Andrew Morton:
 - Some pidns/fork/exec tweaks
 - OCFS2 updates
 - Most of MM - there remain quite a few memcg parts which depend on
   pending core cgroups changes.  Which might have been already merged -
   I'll check tomorrow...
 - Various misc stuff all over the place
 - A few block bits which I never got around to sending to Jens -
   relatively minor things.
 - MAINTAINERS maintenance
 - A small number of lib/ updates
 - checkpatch updates
 - epoll
 - firmware/dmi-scan
 - Some kprobes work for S390
 - drivers/rtc updates
 - hfsplus feature work
 - vmcore feature work
 - rbtree upgrades
 - AOE updates
 - pktcdvd cleanups
 - PPS
 - memstick
 - w1
 - New "inittmpfs" feature, which does the obvious
 - More IPC work from Davidlohr.

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (303 commits)
  lz4: fix compression/decompression signedness mismatch
  ipc: drop ipc_lock_check
  ipc, shm: drop shm_lock_check
  ipc: drop ipc_lock_by_ptr
  ipc, shm: guard against non-existant vma in shmdt(2)
  ipc: document general ipc locking scheme
  ipc,msg: drop msg_unlock
  ipc: rename ids->rw_mutex
  ipc,shm: shorten critical region for shmat
  ipc,shm: cleanup do_shmat pasta
  ipc,shm: shorten critical region for shmctl
  ipc,shm: make shmctl_nolock lockless
  ipc,shm: introduce shmctl_nolock
  ipc: drop ipcctl_pre_down
  ipc,shm: shorten critical region in shmctl_down
  ipc,shm: introduce lockless functions to obtain the ipc object
  initmpfs: use initramfs if rootfstype= or root= specified
  initmpfs: make rootfs use tmpfs when CONFIG_TMPFS enabled
  initmpfs: move rootfs code from fs/ramfs/ to init/
  initmpfs: move bdi setup from init_rootfs to init_ramfs
  ...

11 years agolz4: fix compression/decompression signedness mismatch
Sergey Senozhatsky [Wed, 11 Sep 2013 21:26:32 +0000 (14:26 -0700)]
lz4: fix compression/decompression signedness mismatch

LZ4 compression and decompression functions require different in
signedness input/output parameters: unsigned char for compression and
signed char for decompression.

Change decompression API to require "(const) unsigned char *".

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Kyungsik Lee <kyungsik.lee@lge.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Yann Collet <yann.collet.73@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc: drop ipc_lock_check
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:31 +0000 (14:26 -0700)]
ipc: drop ipc_lock_check

No remaining users, we now use ipc_obtain_object_check().

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Cc: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc, shm: drop shm_lock_check
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:30 +0000 (14:26 -0700)]
ipc, shm: drop shm_lock_check

This function was replaced by a the lockless shm_obtain_object_check(),
and no longer has any users.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Cc: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc: drop ipc_lock_by_ptr
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:29 +0000 (14:26 -0700)]
ipc: drop ipc_lock_by_ptr

After previous cleanups and optimizations, this function is no longer
heavily used and we don't have a good reason to keep it.  Update the few
remaining callers and get rid of it.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Cc: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc, shm: guard against non-existant vma in shmdt(2)
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:28 +0000 (14:26 -0700)]
ipc, shm: guard against non-existant vma in shmdt(2)

When !CONFIG_MMU there's a chance we can derefence a NULL pointer when the
VM area isn't found - check the return value of find_vma().

Also, remove the redundant -EINVAL return: retval is set to the proper
return code and *only* changed to 0, when we actually unmap the segments.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Cc: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc: document general ipc locking scheme
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:26 +0000 (14:26 -0700)]
ipc: document general ipc locking scheme

As suggested by Andrew, add a generic initial locking scheme used
throughout all sysv ipc mechanisms.  Documenting the ids rwsem, how rcu
can be enough to do the initial checks and when to actually acquire the
kern_ipc_perm.lock spinlock.

I found that adding it to util.c was generic enough.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,msg: drop msg_unlock
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:25 +0000 (14:26 -0700)]
ipc,msg: drop msg_unlock

There is only one user left, drop this function and just call
ipc_unlock_object() and rcu_read_unlock().

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc: rename ids->rw_mutex
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:24 +0000 (14:26 -0700)]
ipc: rename ids->rw_mutex

Since in some situations the lock can be shared for readers, we shouldn't
be calling it a mutex, rename it to rwsem.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,shm: shorten critical region for shmat
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:23 +0000 (14:26 -0700)]
ipc,shm: shorten critical region for shmat

Similar to other system calls, acquire the kern_ipc_perm lock after doing
the initial permission and security checks.

[sasha.levin@oracle.com: dont leave do_shmat with rcu lock held]
Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,shm: cleanup do_shmat pasta
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:22 +0000 (14:26 -0700)]
ipc,shm: cleanup do_shmat pasta

Clean up some of the messy do_shmat() spaghetti code, getting rid of
out_free and out_put_dentry labels.  This makes shortening the critical
region of this function in the next patch a little easier to do and read.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,shm: shorten critical region for shmctl
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:21 +0000 (14:26 -0700)]
ipc,shm: shorten critical region for shmctl

With the *_INFO, *_STAT, IPC_RMID and IPC_SET commands already optimized,
deal with the remaining SHM_LOCK and SHM_UNLOCK commands.  Take the
shm_perm lock after doing the initial auditing and security checks.  The
rest of the logic remains unchanged.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,shm: make shmctl_nolock lockless
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:20 +0000 (14:26 -0700)]
ipc,shm: make shmctl_nolock lockless

While the INFO cmd doesn't take the ipc lock, the STAT commands do acquire
it unnecessarily.  We can do the permissions and security checks only
holding the rcu lock.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,shm: introduce shmctl_nolock
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:18 +0000 (14:26 -0700)]
ipc,shm: introduce shmctl_nolock

Similar to semctl and msgctl, when calling msgctl, the *_INFO and *_STAT
commands can be performed without acquiring the ipc object.

Add a shmctl_nolock() function and move the logic of *_INFO and *_STAT out
of msgctl().  Since we are just moving functionality, this change still
takes the lock and it will be properly lockless in the next patch.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc: drop ipcctl_pre_down
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:17 +0000 (14:26 -0700)]
ipc: drop ipcctl_pre_down

Now that sem, msgque and shm, through *_down(), all use the lockless
variant of ipcctl_pre_down(), go ahead and delete it.

[akpm@linux-foundation.org: fix function name in kerneldoc, cleanups]
Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,shm: shorten critical region in shmctl_down
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:16 +0000 (14:26 -0700)]
ipc,shm: shorten critical region in shmctl_down

Instead of holding the ipc lock for the entire function, use the
ipcctl_pre_down_nolock and only acquire the lock for specific commands:
RMID and SET.

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoipc,shm: introduce lockless functions to obtain the ipc object
Davidlohr Bueso [Wed, 11 Sep 2013 21:26:15 +0000 (14:26 -0700)]
ipc,shm: introduce lockless functions to obtain the ipc object

This is the third and final patchset that deals with reducing the amount
of contention we impose on the ipc lock (kern_ipc_perm.lock).  These
changes mostly deal with shared memory, previous work has already been
done for semaphores and message queues:

  http://lkml.org/lkml/2013/3/20/546 (sems)
  http://lkml.org/lkml/2013/5/15/584 (mqueues)

With these patches applied, a custom shm microbenchmark stressing shmctl
doing IPC_STAT with 4 threads a million times, reduces the execution
time by 50%.  A similar run, this time with IPC_SET, reduces the
execution time from 3 mins and 35 secs to 27 seconds.

Patches 1-8: replaces blindly taking the ipc lock for a smarter
combination of rcu and ipc_obtain_object, only acquiring the spinlock
when updating.

Patch 9: renames the ids rw_mutex to rwsem, which is what it already was.

Patch 10: is a trivial mqueue leftover cleanup

Patch 11: adds a brief lock scheme description, requested by Andrew.

This patch:

Add shm_obtain_object() and shm_obtain_object_check(), which will allow us
to get the ipc object without acquiring the lock.  Just as with other
forms of ipc, these functions are basically wrappers around
ipc_obtain_object*().

Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoinitmpfs: use initramfs if rootfstype= or root= specified
Rob Landley [Wed, 11 Sep 2013 21:26:13 +0000 (14:26 -0700)]
initmpfs: use initramfs if rootfstype= or root= specified

Command line option rootfstype=ramfs to obtain old initramfs behavior, and
use ramfs instead of tmpfs for stub when root= defined (for cosmetic
reasons).

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Rob Landley <rob@landley.net>
Cc: Jeff Layton <jlayton@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Stephen Warren <swarren@nvidia.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jim Cromie <jim.cromie@gmail.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoinitmpfs: make rootfs use tmpfs when CONFIG_TMPFS enabled
Rob Landley [Wed, 11 Sep 2013 21:26:12 +0000 (14:26 -0700)]
initmpfs: make rootfs use tmpfs when CONFIG_TMPFS enabled

Conditionally call the appropriate fs_init function and fill_super
functions.  Add a use once guard to shmem_init() to simply succeed on a
second call.

(Note that IS_ENABLED() is a compile time constant so dead code
elimination removes unused function calls when CONFIG_TMPFS is disabled.)

Signed-off-by: Rob Landley <rob@landley.net>
Cc: Jeff Layton <jlayton@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Stephen Warren <swarren@nvidia.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jim Cromie <jim.cromie@gmail.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoinitmpfs: move rootfs code from fs/ramfs/ to init/
Rob Landley [Wed, 11 Sep 2013 21:26:10 +0000 (14:26 -0700)]
initmpfs: move rootfs code from fs/ramfs/ to init/

When the rootfs code was a wrapper around ramfs, having them in the same
file made sense.  Now that it can wrap another filesystem type, move it in
with the init code instead.

This also allows a subsequent patch to access rootfstype= command line
arg.

Signed-off-by: Rob Landley <rob@landley.net>
Cc: Jeff Layton <jlayton@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Stephen Warren <swarren@nvidia.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jim Cromie <jim.cromie@gmail.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoinitmpfs: move bdi setup from init_rootfs to init_ramfs
Rob Landley [Wed, 11 Sep 2013 21:26:08 +0000 (14:26 -0700)]
initmpfs: move bdi setup from init_rootfs to init_ramfs

Even though ramfs hasn't got a backing device, commit e0bf68ddec4f ("mm:
bdi init hooks") added one anyway, and put the initialization in
init_rootfs() since that's the first user, leaving it out of init_ramfs()
to avoid duplication.

But initmpfs uses init_tmpfs() instead, so move the init into the
filesystem's init function, add a "once" guard to prevent duplicate
initialization, and call the filesystem init from rootfs init.

This goes part of the way to allowing ramfs to be built as a module.

[akpm@linux-foundation.org; using bit 1 was odd]
Signed-off-by: Rob Landley <rob@landley.net>
Cc: Jeff Layton <jlayton@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Stephen Warren <swarren@nvidia.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jim Cromie <jim.cromie@gmail.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoinitmpfs: replace MS_NOUSER in initramfs
Rob Landley [Wed, 11 Sep 2013 21:26:06 +0000 (14:26 -0700)]
initmpfs: replace MS_NOUSER in initramfs

Mounting MS_NOUSER prevents --bind mounts from rootfs.  Prevent new rootfs
mounts with a different mechanism that doesn't affect bind mounts.

Signed-off-by: Rob Landley <rob@landley.net>
Cc: Jeff Layton <jlayton@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Stephen Warren <swarren@nvidia.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jim Cromie <jim.cromie@gmail.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agolib/radix-tree.c: make radix_tree_node_alloc() work correctly within interrupt
Jan Kara [Wed, 11 Sep 2013 21:26:05 +0000 (14:26 -0700)]
lib/radix-tree.c: make radix_tree_node_alloc() work correctly within interrupt

With users of radix_tree_preload() run from interrupt (block/blk-ioc.c is
one such possible user), the following race can happen:

radix_tree_preload()
...
radix_tree_insert()
  radix_tree_node_alloc()
    if (rtp->nr) {
      ret = rtp->nodes[rtp->nr - 1];
<interrupt>
...
radix_tree_preload()
...
radix_tree_insert()
  radix_tree_node_alloc()
    if (rtp->nr) {
      ret = rtp->nodes[rtp->nr - 1];

And we give out one radix tree node twice.  That clearly results in radix
tree corruption with different results (usually OOPS) depending on which
two users of radix tree race.

We fix the problem by making radix_tree_node_alloc() always allocate fresh
radix tree nodes when in interrupt.  Using preloading when in interrupt
doesn't make sense since all the allocations have to be atomic anyway and
we cannot steal nodes from process-context users because some users rely
on radix_tree_insert() succeeding after radix_tree_preload().
in_interrupt() check is somewhat ugly but we cannot simply key off passed
gfp_mask as that is acquired from root_gfp_mask() and thus the same for
all preload users.

Another part of the fix is to avoid node preallocation in
radix_tree_preload() when passed gfp_mask doesn't allow waiting.  Again,
preallocation in such case doesn't make sense and when preallocation would
happen in interrupt we could possibly leak some allocated nodes.  However,
some users of radix_tree_preload() require following radix_tree_insert()
to succeed.  To avoid unexpected effects for these users,
radix_tree_preload() only warns if passed gfp mask doesn't allow waiting
and we provide a new function radix_tree_maybe_preload() for those users
which get different gfp mask from different call sites and which are
prepared to handle radix_tree_insert() failure.

Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <jaxboe@fusionio.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agodrivers/w1/masters/mxc_w1.c: remove unnecessary platform_set_drvdata()
Jingoo Han [Wed, 11 Sep 2013 21:26:04 +0000 (14:26 -0700)]
drivers/w1/masters/mxc_w1.c: remove unnecessary platform_set_drvdata()

The driver core clears the driver data to NULL after device_release or on
probe failure.  Thus, it is not needed to manually clear the device driver
data to NULL.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Greg KH <greg@kroah.com>
Acked-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agodrivers/w1/w1.c: replace strict_strtol() with kstrtol()
Jingoo Han [Wed, 11 Sep 2013 21:26:03 +0000 (14:26 -0700)]
drivers/w1/w1.c: replace strict_strtol() with kstrtol()

The usage of strict_strtol() is not preferred, because strict_strtol() is
obsolete.  Thus, kstrtol() should be used.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomemstick: add support for legacy memorysticks
Maxim Levitsky [Wed, 11 Sep 2013 21:26:02 +0000 (14:26 -0700)]
memstick: add support for legacy memorysticks

Based partially on MS standard spec quotes from Alex Dubov.

As any code that works with user data this driver isn't recommended to use
to write cards that contain valuable data.

It tries its best though to avoid data corruption and possible damage to
the card.

Tested on MS DUO 64 MB card on Ricoh R592 card reader.

Signed-off-by: Maxim Levitsky <maximlevitsky@gmail.com>
Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Alex Dubov <oakad@yahoo.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agodrivers/memstick/host/rtsx_pci_ms.c: remove unnecessary platform_set_drvdata()
Jingoo Han [Wed, 11 Sep 2013 21:26:01 +0000 (14:26 -0700)]
drivers/memstick/host/rtsx_pci_ms.c: remove unnecessary platform_set_drvdata()

The driver core clears the driver data to NULL after device_release or on
probe failure.  Thus, it is not needed to manually clear the device driver
data to NULL.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Cc: Maxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agodrivers/pps/clients/pps-gpio.c: remove unnecessary platform_set_drvdata()
Jingoo Han [Wed, 11 Sep 2013 21:26:00 +0000 (14:26 -0700)]
drivers/pps/clients/pps-gpio.c: remove unnecessary platform_set_drvdata()

The driver core clears the driver data to NULL after device_release or on
probe failure.  Thus, it is not needed to manually clear the device driver
data to NULL.

Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Cc: Rodolfo Giometti <giometti@enneenne.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: fix defective misuses of pkt_<level>
Joe Perches [Wed, 11 Sep 2013 21:25:59 +0000 (14:25 -0700)]
pktcdvd: fix defective misuses of pkt_<level>

Fix thinkos where pkt_<level> needs a valid pktcdvd_device * and the
pointer is known to be NULL.

Signed-off-by: Joe Perches <joe@perches.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com> (go smatch!)
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: add struct pktcdvd_device * to pkt_dump_sense()
Joe Perches [Wed, 11 Sep 2013 21:25:58 +0000 (14:25 -0700)]
pktcdvd: add struct pktcdvd_device * to pkt_dump_sense()

Allow the device name to be emitted with pkt_err when logging the sense
data.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: convert pr_info to pkt_info
Joe Perches [Wed, 11 Sep 2013 21:25:57 +0000 (14:25 -0700)]
pktcdvd: convert pr_info to pkt_info

Add a new pkt_info macro to prefix the name to the logging output.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: convert pr_notice to pkt_notice
Joe Perches [Wed, 11 Sep 2013 21:25:56 +0000 (14:25 -0700)]
pktcdvd: convert pr_notice to pkt_notice

Add a new pkt_notice macro to prefix the name to the logging output.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: add struct pktcdvd_device.name to pr_err logging where possible
Joe Perches [Wed, 11 Sep 2013 21:25:55 +0000 (14:25 -0700)]
pktcdvd: add struct pktcdvd_device.name to pr_err logging where possible

Add a new pkt_err macro to prefix the name to the logging output.  Convert
pr_err where there is a non-null struct pktcdvd_device.

Includes improvements from Andy Shevchenko.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Andy Shevchenko <andy.shevchenko@gmail.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: add struct pktcdvd_device * to pkt_dbg
Joe Perches [Wed, 11 Sep 2013 21:25:54 +0000 (14:25 -0700)]
pktcdvd: add struct pktcdvd_device * to pkt_dbg

Add pd->name to output for these debugging messages.

Remove normally compiled out pkt_dbg(2, ...) function entry tracing
equivalents as it's better done via the function tracer.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: consolidate DPRINTK and VPRINTK macros
Joe Perches [Wed, 11 Sep 2013 21:25:53 +0000 (14:25 -0700)]
pktcdvd: consolidate DPRINTK and VPRINTK macros

Use the more common pkt_dbg(level, fmt, ...) form.

These messages are emitted at KERN_NOTICE.

Always emit function name with pkt_dbg(2, ...) uses and remove the
sometimes abbreviated embedded function name.

This form always verifies the format and arguments.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: convert printk to pr_<level>
Joe Perches [Wed, 11 Sep 2013 21:25:52 +0000 (14:25 -0700)]
pktcdvd: convert printk to pr_<level>

Use a more current logging style and add messages levels to the logging
messages.

Simplify pkt_dump_sense by using %*ph and adding a simple function to emit
the sense string.

Includes improvements from Andy Shevchenko and Dan Carpenter.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Andy Shevchenko <andy.shevchenko@gmail.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopktcdvd: convert ZONE macro to static function get_zone()
Joe Perches [Wed, 11 Sep 2013 21:25:51 +0000 (14:25 -0700)]
pktcdvd: convert ZONE macro to static function get_zone()

Macros should be converted to functions where feasible to
verify arguments and the like.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopanic: call panic handlers before kmsg_dump
Kees Cook [Wed, 11 Sep 2013 21:25:49 +0000 (14:25 -0700)]
panic: call panic handlers before kmsg_dump

Since the panic handlers may produce additional information (via printk)
for the kernel log, it should be reported as part of the panic output
saved by kmsg_dump().  Without this re-ordering, nothing that adds
information to a panic will show up in pstore's view when kmsg_dump runs,
and is therefore not visible to crash reporting tools that examine pstore
output.

Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Anton Vorontsov <anton@enomsg.org>
Cc: Colin Cross <ccross@android.com>
Acked-by: Tony Luck <tony.luck@intel.com>
Cc: Stephen Boyd <sboyd@codeaurora.org>
Cc: Vikram Mulukutla <markivx@codeaurora.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaffs: use loff_t in affs_truncate()
Dan Carpenter [Wed, 11 Sep 2013 21:25:48 +0000 (14:25 -0700)]
affs: use loff_t in affs_truncate()

It seems pretty unlikely that AFFS supports files over 4GB but we may as
well leave use loff_t just for cleanness sake instead of truncating it to
32 bits.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Marco Stornelli <marco.stornelli@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: remove do-nothing NAME="%k" term from example udev rules
Ed Cashin [Wed, 11 Sep 2013 21:25:47 +0000 (14:25 -0700)]
aoe: remove do-nothing NAME="%k" term from example udev rules

When the example udev rules in the documentation are used without
modification, warnings like the one shown below appear in the system logs:

  /var/log/messages:Aug 22 11:09:11 kung udevd[445]: NAME="%k" \
    is superfluous and breaks kernel supplied names, please remove \
    it from /etc/udev/rules.d/60-aoe.rules:26

Removing the term does not cause any problems with the creation of the
special character and block device nodes.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: do not BUG if memory pressure prevented debugfs file creation
Ed Cashin [Wed, 11 Sep 2013 21:25:46 +0000 (14:25 -0700)]
aoe: do not BUG if memory pressure prevented debugfs file creation

If the system has trouble allocating memory for the creation of the aoe
debugfs directory or of a file inside it, the debugfs member of an aoedev
can be NULL.

Do not treat a NULL debugfs pointer as a BUG on aoedev shutdown, avoiding
the user impact of an unecessary panic.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: suppress compiler warnings
Andy Shevchenko [Wed, 11 Sep 2013 21:25:45 +0000 (14:25 -0700)]
aoe: suppress compiler warnings

This patch fixes following compiler warnings:

  drivers/block/aoe/aoecmd.c: In function `aoecmd_ata_rw':
  drivers/block/aoe/aoecmd.c:383:17: warning: variable `t' set but not used [-Wunused-but-set-variable]
    struct aoetgt *t;
                   ^
  drivers/block/aoe/aoecmd.c: In function `resend':
  drivers/block/aoe/aoecmd.c:488:21: warning: variable `ah' set but not used [-Wunused-but-set-variable]
    struct aoe_atahdr *ah;
                       ^

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: remove custom implementation of kbasename()
Andy Shevchenko [Wed, 11 Sep 2013 21:25:44 +0000 (14:25 -0700)]
aoe: remove custom implementation of kbasename()

In the kernel we have a nice helper that may be used here. This patch
substitutes the custom implementation by the native function call.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: update internal version number to 85
Ed Cashin [Wed, 11 Sep 2013 21:25:44 +0000 (14:25 -0700)]
aoe: update internal version number to 85

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: update copyright date
Ed Cashin [Wed, 11 Sep 2013 21:25:43 +0000 (14:25 -0700)]
aoe: update copyright date

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: fill in per-AoE-target information for debugfs file
Ed Cashin [Wed, 11 Sep 2013 21:25:42 +0000 (14:25 -0700)]
aoe: fill in per-AoE-target information for debugfs file

This information is presented in a compact format that has evolved for
easy routine scanning by expert humans, mostly developers and support
technicians helping to troubleshoot or test AoE-based systems.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: provide file operations for debugfs files
Ed Cashin [Wed, 11 Sep 2013 21:25:41 +0000 (14:25 -0700)]
aoe: provide file operations for debugfs files

The place holder in the file contents is filled out in the following
patch.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: add AoE-target files to debugfs
Ed Cashin [Wed, 11 Sep 2013 21:25:40 +0000 (14:25 -0700)]
aoe: add AoE-target files to debugfs

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoaoe: create and destroy debugfs directory for aoe
Ed Cashin [Wed, 11 Sep 2013 21:25:39 +0000 (14:25 -0700)]
aoe: create and destroy debugfs directory for aoe

This series adds the debugging information that the coraid.com-distributed
aoe driver exports via sysfs, but instead of sysfs, it uses debugfs.

With these patches applied, even without AoE targets on the network, KEDR
reports new possible memory leaks, but these are from callers outside the
aoe driver that have used aoe_devnode to get the name of the character
devices through the aoe_class->devnode callback, and I believe they're
responsible for freeing that memory.

This patch:

Create and destroy the debugfs directory.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agomm/zswap: use postorder iteration when destroying rbtree
Cody P Schafer [Wed, 11 Sep 2013 21:25:33 +0000 (14:25 -0700)]
mm/zswap: use postorder iteration when destroying rbtree

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Reviewed-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: David Woodhouse <David.Woodhouse@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agorbtree: allow tests to run as builtin
Cody P Schafer [Wed, 11 Sep 2013 21:25:19 +0000 (14:25 -0700)]
rbtree: allow tests to run as builtin

No reason require rbtree test code to be a module, allow it to be builtin
(streamlines my development process)

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Reviewed-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: David Woodhouse <David.Woodhouse@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agorbtree_test: add test for postorder iteration
Cody P Schafer [Wed, 11 Sep 2013 21:25:17 +0000 (14:25 -0700)]
rbtree_test: add test for postorder iteration

Just check that we examine all nodes in the tree for the postorder
iteration.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Reviewed-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: David Woodhouse <David.Woodhouse@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agorbtree: add rbtree_postorder_for_each_entry_safe() helper
Cody P Schafer [Wed, 11 Sep 2013 21:25:11 +0000 (14:25 -0700)]
rbtree: add rbtree_postorder_for_each_entry_safe() helper

Because deletion (of the entire tree) is a relatively common use of the
rbtree_postorder iteration, and because doing it safely means fiddling
with temporary storage, provide a helper to simplify postorder rbtree
iteration.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Reviewed-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: David Woodhouse <David.Woodhouse@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agorbtree: add postorder iteration functions
Cody P Schafer [Wed, 11 Sep 2013 21:25:10 +0000 (14:25 -0700)]
rbtree: add postorder iteration functions

Postorder iteration yields all of a node's children prior to yielding the
node itself, and this particular implementation also avoids examining the
leaf links in a node after that node has been yielded.

In what I expect will be its most common usage, postorder iteration allows
the deletion of every node in an rbtree without modifying the rbtree nodes
(no _requirement_ that they be nulled) while avoiding referencing child
nodes after they have been "deleted" (most commonly, freed).

I have only updated zswap to use this functionality at this point, but
numerous bits of code (most notably in the filesystem drivers) use a hand
rolled postorder iteration that NULLs child links as it traverses the
tree.  Each of those instances could be replaced with this common
implementation.

1 & 2 add rbtree postorder iteration functions.
3 adds testing of the iteration to the rbtree runtime tests
4 allows building the rbtree runtime tests as builtins
5 updates zswap.

This patch:

Add postorder iteration functions for rbtree.  These are useful for safely
freeing an entire rbtree without modifying the tree at all.

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Reviewed-by: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: David Woodhouse <David.Woodhouse@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoblock/partitions/efi.c: consistently use pr_foo()
Andrew Morton [Wed, 11 Sep 2013 21:25:04 +0000 (14:25 -0700)]
block/partitions/efi.c: consistently use pr_foo()

Cc: Davidlohr Bueso <davidlohr@hp.com>
Cc: Karel Zak <kzak@redhat.com>
Cc: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: some style cleanups
Davidlohr Bueso [Wed, 11 Sep 2013 21:25:03 +0000 (14:25 -0700)]
partitions/efi: some style cleanups

Trivial coding style cleanups - still plenty left.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: delete annoying emacs style comments
Davidlohr Bueso [Wed, 11 Sep 2013 21:25:02 +0000 (14:25 -0700)]
partitions/efi: delete annoying emacs style comments

I love emacs, but these settings for coding style are annoying when trying
to open the efi.h file.  More important, we already have checkpatch for
that.

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: compare first and last usable LBAs
Davidlohr Bueso [Wed, 11 Sep 2013 21:25:01 +0000 (14:25 -0700)]
partitions/efi: compare first and last usable LBAs

When verifying GPT header integrity, make sure that first usable LBA is
smaller than last usable LBA.

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: account for pmbr size in lba
Davidlohr Bueso [Wed, 11 Sep 2013 21:25:00 +0000 (14:25 -0700)]
partitions/efi: account for pmbr size in lba

The partition that has the 0xEE (GPT protective), must have the size in
lba field set to the lesser of the size of the disk minus one or
0xFFFFFFFF for larger disks.

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: detect hybrid MBRs
Davidlohr Bueso [Wed, 11 Sep 2013 21:24:58 +0000 (14:24 -0700)]
partitions/efi: detect hybrid MBRs

One of the biggest problems with GPT is compatibility with older, non-GPT
systems.  The problem is addressed by creating hybrid mbrs, an extension,
or variant, of the traditional protective mbr.  This contains, apart from
the 0xEE partition, up three additional primary partitions that point to
the same space marked by up to three GPT partitions.  The result is that
legacy OSs can see the three required MBR partitions and at the same time
ignore the GPT-aware partitions that protect the GPT structures.

While hybrid MBRs are hacks, workarounds and simply not part of the GPT
standard, they do exist and we have no way around them.  For instance, by
default, OSX creates a hybrid scheme when using multi-OS booting.

In order for Linux to properly discover protective MBRs, it must be made
aware of devices that have hybrid MBRs.  No functionality is changed by
this patch, just a debug message informing the user of the MBR scheme that
is being used.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: do not require gpt partition to begin at sector 1
Davidlohr Bueso [Wed, 11 Sep 2013 21:24:57 +0000 (14:24 -0700)]
partitions/efi: do not require gpt partition to begin at sector 1

When detecting a valid protective MBR, the Linux kernel isn't picky about
the partition (1-4) the 0xEE is at, but, unlike other operating systems,
it does require it to begin at the second sector (sector 1).  This check,
apart from it not being enforced by UEFI, and causing Linux to potentially
fail to detect any *valid* partitions on the disk, can present problems
when dealing with hybrid MBRs[1].

For compatibility reasons, if the first partition is hybridized, the 0xEE
partition must be small enough to ensure that it only protects the GPT
data structures - as opposed to the the whole disk in a protective MBR.
This problem is very well described by Rod Smith[1]: where MBR-only
partitioning programs (such as older versions of fdisk) can see some of
the disk space as unallocated, thus loosing the purpose of the 0xEE
partition's protection of GPT data structures.

By dropping this check, this patch enables Linux to be more flexible when
probing for GPT disklabels.

[1] http://www.rodsbooks.com/gdisk/hybrid.html#reactions

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: check pmbr record's starting lba
Davidlohr Bueso [Wed, 11 Sep 2013 21:24:56 +0000 (14:24 -0700)]
partitions/efi: check pmbr record's starting lba

Per the UEFI Specs 2.4, June 2013, the starting lba of the partition that
has the EFI GPT (0xEE) must be set to 0x00000001 - this is obviously the
LBA of the GPT Partition Header.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agopartitions/efi: use lba-aware partition records
Davidlohr Bueso [Wed, 11 Sep 2013 21:24:55 +0000 (14:24 -0700)]
partitions/efi: use lba-aware partition records

The kernel's GPT implementation currently uses the generic 'struct
partition' type for dealing with legacy MBR partition records.  While this
is is useful for disklabels that we designed for CHS addressing, such as
msdos, it doesn't adapt well to newer standards that use LBA instead, such
as GUID partition tables.  Furthermore, these generic partition structures
do not have all the required fields to properly follow the UEFI specs.

While a CHS address can be translated to LBA, it's much simpler and
cleaner to just replace the partition type.  This patch adds a new
'gpt_record' type that is fully compliant with EFI and will allow, in the
next patches, to add more checks to properly verify a protective MBR,
which is paramount to probing a device that makes use of GPT.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Karel Zak <kzak@redhat.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agos390/vmcore: use vmcore for zfcpdump
Michael Holzheu [Wed, 11 Sep 2013 21:24:54 +0000 (14:24 -0700)]
s390/vmcore: use vmcore for zfcpdump

Modify the s390 copy_oldmem_page() and remap_oldmem_pfn_range() function
for zfcpdump to read from the HSA memory if memory below HSA_SIZE bytes is
requested.  Otherwise real memory is used.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Jan Willeke <willeke@de.ibm.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agovmcore: enable /proc/vmcore mmap for s390
Michael Holzheu [Wed, 11 Sep 2013 21:24:53 +0000 (14:24 -0700)]
vmcore: enable /proc/vmcore mmap for s390

The patch "s390/vmcore: Implement remap_oldmem_pfn_range for s390" allows
now to use mmap also on s390.

So enable mmap for s390 again.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Jan Willeke <willeke@de.ibm.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agos390/vmcore: implement remap_oldmem_pfn_range for s390
Jan Willeke [Wed, 11 Sep 2013 21:24:52 +0000 (14:24 -0700)]
s390/vmcore: implement remap_oldmem_pfn_range for s390

Introduce the s390 specific way to map pages from oldmem.  The memory area
below OLDMEM_SIZE is mapped with offset OLDMEM_BASE.  The other old memory
is mapped directly.

Signed-off-by: Jan Willeke <willeke@de.ibm.com>
Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agovmcore: introduce remap_oldmem_pfn_range()
Michael Holzheu [Wed, 11 Sep 2013 21:24:51 +0000 (14:24 -0700)]
vmcore: introduce remap_oldmem_pfn_range()

For zfcpdump we can't map the HSA storage because it is only available via
a read interface.  Therefore, for the new vmcore mmap feature we have
introduce a new mechanism to create mappings on demand.

This patch introduces a new architecture function remap_oldmem_pfn_range()
that should be used to create mappings with remap_pfn_range() for oldmem
areas that can be directly mapped.  For zfcpdump this is everything
besides of the HSA memory.  For the areas that are not mapped by
remap_oldmem_pfn_range() a generic vmcore a new generic vmcore fault
handler mmap_vmcore_fault() is called.

This handler works as follows:

* Get already available or new page from page cache (find_or_create_page)
* Check if /proc/vmcore page is filled with data (PageUptodate)
* If yes:
  Return that page
* If no:
  Fill page using __vmcore_read(), set PageUptodate, and return page

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Jan Willeke <willeke@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agos390/vmcore: use ELF header in new memory feature
Michael Holzheu [Wed, 11 Sep 2013 21:24:50 +0000 (14:24 -0700)]
s390/vmcore: use ELF header in new memory feature

Exchange the old relocate mechanism with the new arch function call
override mechanism that allows to create the ELF core header in the 2nd
kernel.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Jan Willeke <willeke@de.ibm.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agovmcore: introduce ELF header in new memory feature
Michael Holzheu [Wed, 11 Sep 2013 21:24:49 +0000 (14:24 -0700)]
vmcore: introduce ELF header in new memory feature

For s390 we want to use /proc/vmcore for our SCSI stand-alone dump
(zfcpdump).  We have support where the first HSA_SIZE bytes are saved into
a hypervisor owned memory area (HSA) before the kdump kernel is booted.
When the kdump kernel starts, it is restricted to use only HSA_SIZE bytes.

The advantages of this mechanism are:

 * No crashkernel memory has to be defined in the old kernel.
 * Early boot problems (before kexec_load has been done) can be dumped
 * Non-Linux systems can be dumped.

We modify the s390 copy_oldmem_page() function to read from the HSA memory
if memory below HSA_SIZE bytes is requested.

Since we cannot use the kexec tool to load the kernel in this scenario,
we have to build the ELF header in the 2nd (kdump/new) kernel.

So with the following patch set we would like to introduce the new
function that the ELF header for /proc/vmcore can be created in the 2nd
kernel memory.

The following steps are done during zfcpdump execution:

1.  Production system crashes
2.  User boots a SCSI disk that has been prepared with the zfcpdump tool
3.  Hypervisor saves CPU state of boot CPU and HSA_SIZE bytes of memory into HSA
4.  Boot loader loads kernel into low memory area
5.  Kernel boots and uses only HSA_SIZE bytes of memory
6.  Kernel saves registers of non-boot CPUs
7.  Kernel does memory detection for dump memory map
8.  Kernel creates ELF header for /proc/vmcore
9.  /proc/vmcore uses this header for initialization
10. The zfcpdump user space reads /proc/vmcore to write dump to SCSI disk
    - copy_oldmem_page() copies from HSA for memory below HSA_SIZE
    - copy_oldmem_page() copies from real memory for memory above HSA_SIZE

Currently for s390 we create the ELF core header in the 2nd kernel with a
small trick.  We relocate the addresses in the ELF header in a way that
for the /proc/vmcore code it seems to be in the 1st kernel (old) memory
and the read_from_oldmem() returns the correct data.  This allows the
/proc/vmcore code to use the ELF header in the 2nd kernel.

This patch:

Exchange the old mechanism with the new and much cleaner function call
override feature that now offcially allows to create the ELF core header
in the 2nd kernel.

To use the new feature the following function have to be defined
by the architecture backend code to read from new memory:

 * elfcorehdr_alloc: Allocate ELF header
 * elfcorehdr_free: Free the memory of the ELF header
 * elfcorehdr_read: Read from ELF header
 * elfcorehdr_read_notes: Read from ELF notes

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Jan Willeke <willeke@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agokexec: remove unnecessary return
Xishi Qiu [Wed, 11 Sep 2013 21:24:47 +0000 (14:24 -0700)]
kexec: remove unnecessary return

Code can not run here forever, so remove the unnecessary return.

Signed-off-by: Xishi Qiu <qiuxishi@huawei.com>
Suggested-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Reviewed-by: Simon Horman <horms@verge.net.au>
Reviewed-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: cleanup the error handling in search_binary_handler()
Oleg Nesterov [Wed, 11 Sep 2013 21:24:46 +0000 (14:24 -0700)]
exec: cleanup the error handling in search_binary_handler()

The error hanling and ret-from-loop look confusing and inconsistent.

- "retval >= 0" simply returns

- "!bprm->file" returns too but with read_unlock() because
   binfmt_lock was already re-acquired

- "retval != -ENOEXEC || bprm->mm == NULL" does "break" and
  relies on the same check after the main loop

Consolidate these checks into a single if/return statement.

need_retry still checks "retval == -ENOEXEC", but this and -ENOENT before
the main loop are not needed.  This is only for pathological and
impossible list_empty(&formats) case.

It is not clear why do we check "bprm->mm == NULL", probably this
should be removed.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: don't retry if request_module() fails
Oleg Nesterov [Wed, 11 Sep 2013 21:24:45 +0000 (14:24 -0700)]
exec: don't retry if request_module() fails

A separate one-liner for better documentation.

It doesn't make sense to retry if request_module() fails to exec
/sbin/modprobe, add the additional "request_module() < 0" check.

However, this logic still doesn't look exactly right:

1. It would be better to check "request_module() != 0", the user
   space modprobe process should report the correct exit code.
   But I didn't dare to add the user-visible change.

2. The whole ENOEXEC logic looks suboptimal. Suppose that we try
   to exec a "#!path-to-unsupported-binary" script. In this case
   request_module() + "retry" will be done twice: first by the
   "depth == 1" code, and then again by the "depth == 0" caller
   which doesn't make sense.

3. And note that in the case above bprm->buf was already changed
   by load_script()->prepare_binprm(), so this looks even more
   ugly.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: cleanup the CONFIG_MODULES logic
Oleg Nesterov [Wed, 11 Sep 2013 21:24:44 +0000 (14:24 -0700)]
exec: cleanup the CONFIG_MODULES logic

search_binary_handler() uses "for (try=0; try<2; try++)" to avoid "goto"
but the code looks too complicated and horrible imho.  We still need to
check "try == 0" before request_module() and add the additional "break"
for !CONFIG_MODULES case.

Kill this loop and use a simple "bool need_retry" + "goto retry".  The
code looks much simpler and we do not even need ifdef's, gcc can optimize
out the "if (need_retry)" block if !IS_ENABLED().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: kill ->load_binary != NULL check in search_binary_handler()
Oleg Nesterov [Wed, 11 Sep 2013 21:24:42 +0000 (14:24 -0700)]
exec: kill ->load_binary != NULL check in search_binary_handler()

search_binary_handler() checks ->load_binary != NULL for no reason, this
method should be always defined.  Turn this check into WARN_ON() and move
it into __register_binfmt().

Also, kill the function pointer.  The current code looks confusing, as if
->load_binary can go away after read_unlock(&binfmt_lock).  But we rely on
module_get(fmt->module), this fmt can't be changed or unregistered,
otherwise this code is buggy anyway.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: move allow_write_access/fput to exec_binprm()
Oleg Nesterov [Wed, 11 Sep 2013 21:24:41 +0000 (14:24 -0700)]
exec: move allow_write_access/fput to exec_binprm()

When search_binary_handler() succeeds it does allow_write_access() and
fput(), then it clears bprm->file to ensure the caller will not do the
same.

We can simply move this code to exec_binprm() which is called only once.
In fact we could move this to free_bprm() and remove the same code in
do_execve_common's error path.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: proc_exec_connector() should be called only once
Oleg Nesterov [Wed, 11 Sep 2013 21:24:40 +0000 (14:24 -0700)]
exec: proc_exec_connector() should be called only once

A separate one-liner with the minor fix.

PROC_EVENT_EXEC reports the "exec" event, but this message is sent at
least twice if search_binary_handler() is called by ->load_binary()
recursively, say, load_script().

Move it to exec_binprm(), this is "depth == 0" code too.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: kill "int depth" in search_binary_handler()
Oleg Nesterov [Wed, 11 Sep 2013 21:24:39 +0000 (14:24 -0700)]
exec: kill "int depth" in search_binary_handler()

Nobody except search_binary_handler() should touch ->recursion_depth, "int
depth" buys nothing but complicates the code, kill it.

Probably we should also kill "fn" and the !NULL check, ->load_binary
should be always defined.  And it can not go away after read_unlock() or
this code is buggy anyway.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoexec: introduce exec_binprm() for "depth == 0" code
Oleg Nesterov [Wed, 11 Sep 2013 21:24:38 +0000 (14:24 -0700)]
exec: introduce exec_binprm() for "depth == 0" code

task_pid_nr_ns() and trace/ptrace code in the middle of the recursive
search_binary_handler() looks confusing and imho annoying.  We only need
this code if "depth == 0", lets add a simple helper which calls
search_binary_handler() and does trace_sched_process_exec() +
ptrace_event().

The patch also moves the setting of task->did_exec, we need to do this
only once.

Note: we can kill either task->did_exec or PF_FORKNOEXEC.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Zach Levis <zml@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoproc: make proc_fd_permission() thread-friendly
Oleg Nesterov [Wed, 11 Sep 2013 21:24:37 +0000 (14:24 -0700)]
proc: make proc_fd_permission() thread-friendly

proc_fd_permission() says "process can still access /proc/self/fd after it
has executed a setuid()", but the "task_pid() = proc_pid() check only
helps if the task is group leader, /proc/self points to
/proc/<leader-pid>.

Change this check to use task_tgid() so that the whole thread group can
access its /proc/self/fd or /proc/<tid-of-sub-thread>/fd.

Notes:
- CLONE_THREAD does not require CLONE_FILES so task->files
  can differ, but I don't think this can lead to any security
  problem. And this matches same_thread_group() in
  __ptrace_may_access().

- /proc/self should probably point to /proc/<thread-tid>, but
  it is too late to change the rules. Perhaps it makes sense
  to add /proc/thread though.

Test-case:

void *tfunc(void *arg)
{
assert(opendir("/proc/self/fd"));
return NULL;
}

int main(void)
{
pthread_t t;
pthread_create(&t, NULL, tfunc, NULL);
pthread_join(t, NULL);
return 0;
}

fails if, say, this executable is not readable and suid_dumpable = 0.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs/proc/task_mmu.c: check the return value of mpol_to_str()
Chen Gang [Wed, 11 Sep 2013 21:24:35 +0000 (14:24 -0700)]
fs/proc/task_mmu.c: check the return value of mpol_to_str()

mpol_to_str() may fail, and not fill the buffer (e.g. -EINVAL), so need
check about it, or buffer may not be zero based, and next seq_printf()
will cause issue.

The failure return need after mpol_cond_put() to match get_vma_policy().

Signed-off-by: Chen Gang <gang.chen@asianux.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoDocumentation/filesystems/proc.txt: fix mistake in the description of Committed_AS
Minto Joseph [Wed, 11 Sep 2013 21:24:35 +0000 (14:24 -0700)]
Documentation/filesystems/proc.txt: fix mistake in the description of Committed_AS

Fix mistake in the description of Committed_AS in kernel documentation.

Signed-off-by: Minto Joseph <mvaliyav@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agofs/file_table.c:fput(): make comment more truthful
Andrew Morton [Wed, 11 Sep 2013 21:24:34 +0000 (14:24 -0700)]
fs/file_table.c:fput(): make comment more truthful

Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agocoredump: add new %P variable in core_pattern
Stéphane Graber [Wed, 11 Sep 2013 21:24:32 +0000 (14:24 -0700)]
coredump: add new %P variable in core_pattern

Add a new %P variable to be used in core_pattern.  This variable contains
the global PID (PID in the init namespace) as %p contains the PID in the
current namespace which isn't always what we want.

The main use for this is to make it easier to handle crashes that happened
within a container.  With that new variables it's possible to have the
crashes dumped into the container or forwarded to the host with the right
PID (from the host's point of view).

Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>
Reported-by: Hans Feldt <hans.feldt@ericsson.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andy Whitcroft <apw@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years ago__ptrace_may_access() should not deny sub-threads
Mark Grondona [Wed, 11 Sep 2013 21:24:31 +0000 (14:24 -0700)]
__ptrace_may_access() should not deny sub-threads

__ptrace_may_access() checks get_dumpable/ptrace_has_cap/etc if task !=
current, this can can lead to surprising results.

For example, a sub-thread can't readlink("/proc/self/exe") if the
executable is not readable.  setup_new_exec()->would_dump() notices that
inode_permission(MAY_READ) fails and then it does
set_dumpable(suid_dumpable).  After that get_dumpable() fails.

(It is not clear why proc_pid_readlink() checks get_dumpable(), perhaps we
could add PTRACE_MODE_NODUMPABLE)

Change __ptrace_may_access() to use same_thread_group() instead of "task
== current".  Any security check is pointless when the tasks share the
same ->mm.

Signed-off-by: Mark Grondona <mgrondona@llnl.gov>
Signed-off-by: Ben Woodard <woodard@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agohfsplus: integrate POSIX ACLs support into driver
Vyacheslav Dubeyko [Wed, 11 Sep 2013 21:24:30 +0000 (14:24 -0700)]
hfsplus: integrate POSIX ACLs support into driver

Integrate implemented POSIX ACLs support into hfsplus driver.

Signed-off-by: Vyacheslav Dubeyko <slava@dubeyko.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hin-Tak Leung <htl10@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agohfsplus: implement POSIX ACLs support
Vyacheslav Dubeyko [Wed, 11 Sep 2013 21:24:29 +0000 (14:24 -0700)]
hfsplus: implement POSIX ACLs support

Implement POSIX ACLs support in hfsplus driver.

Signed-off-by: Vyacheslav Dubeyko <slava@dubeyko.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hin-Tak Leung <htl10@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agohfsplus: add necessary declarations for POSIX ACLs support
Vyacheslav Dubeyko [Wed, 11 Sep 2013 21:24:28 +0000 (14:24 -0700)]
hfsplus: add necessary declarations for POSIX ACLs support

This patchset implements POSIX ACLs support in hfsplus driver.

Mac OS X beginning with version 10.4 ("Tiger") support NFSv4 ACLs, which
are part of the NFSv4 standard.  HFS+ stores ACLs in the form of
specially named extended attributes (com.apple.system.Security).

But this patchset doesn't use "com.apple.system.Security" extended
attributes.  It implements support of POSIX ACLs in the form of extended
attributes with names "system.posix_acl_access" and
"system.posix_acl_default".  These xattrs are treated only under Linux.
POSIX ACLs doesn't mean something under Mac OS X.  Thereby, this patch
set provides opportunity to use POSIX ACLs under Linux on HFS+
filesystem.

This patch:

Add CONFIG_HFSPLUS_FS_POSIX_ACL kernel configuration option, DBG_ACL_MOD
debugging flag and acl.h file with declaration of essential functions
for support POSIX ACLs in hfsplus driver.

Signed-off-by: Vyacheslav Dubeyko <slava@dubeyko.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Hin-Tak Leung <htl10@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agortc: simplify devm_request_mem_region/devm_ioremap
Julia Lawall [Wed, 11 Sep 2013 21:24:27 +0000 (14:24 -0700)]
rtc: simplify devm_request_mem_region/devm_ioremap

Convert the composition of devm_request_mem_region and devm_ioremap to a
single call to devm_ioremap_resource.  The associated call to
platform_get_resource is also simplified and moved next to the new call
to devm_ioremap_resource.

This was done using a combination of the semantic patches
devm_ioremap_resource.cocci and devm_request_and_ioremap.cocci, found in
the scripts/coccinelle/api directory.

In rtc-lpc32xx.c and rtc-mv.c, the local variable size is no longer needed.

In rtc-ds1511.c the size field of the local structure is not useful any
more, and is deleted.

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agodrivers/rtc/rtc-ds1742.c: report to RTC core if retrieved time is invalid
Alexander Shiyan [Wed, 11 Sep 2013 21:24:26 +0000 (14:24 -0700)]
drivers/rtc/rtc-ds1742.c: report to RTC core if retrieved time is invalid

Let RTC core decide if the retrieved time is invalid, instead of
processing errors in the driver.

Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Cc: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>