GitHub/LineageOS/android_kernel_motorola_exynos9610.git
11 years agocgroup: remove subsystem files when remounting cgroup
Gao feng [Mon, 3 Dec 2012 01:28:18 +0000 (09:28 +0800)]
cgroup: remove subsystem files when remounting cgroup

cgroup_clear_directroy is called by cgroup_d_remove_dir
and cgroup_remount.

when we call cgroup_remount to remount the cgroup,the subsystem
may be unlinked from cgroupfs_root->subsys_list in rebind_subsystem,this
subsystem's files will not be removed in cgroup_clear_directroy.
And the system will panic when we try to access these files.

this patch removes subsystems's files before rebind_subsystems,
if rebind_subsystems failed, repopulate these removed files.

With help from Tejun.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: use cgroup_addrm_files() in cgroup_clear_directory()
Gao feng [Fri, 30 Nov 2012 16:21:28 +0000 (00:21 +0800)]
cgroup: use cgroup_addrm_files() in cgroup_clear_directory()

cgroup_clear_directory() incorrectly invokes cgroup_rm_file() on each
cftset of the target subsystems, which only removes the first file of
each set.  This leaves dangling files after subsystems are removed
from a cgroup root via remount.

Use cgroup_addrm_files() to remove all files of target subsystems.

tj: Move cgroup_addrm_files() prototype decl upwards next to other
    global declarations.  Commit message updated.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: warn about broken hierarchies only after css_online
Glauber Costa [Fri, 30 Nov 2012 13:31:23 +0000 (17:31 +0400)]
cgroup: warn about broken hierarchies only after css_online

If everything goes right, it shouldn't really matter if we are spitting
this warning after css_alloc or css_online. If we fail between then,
there are some ill cases where we would previously see the message and
now we won't (like if the files fail to be created).

I believe it really shouldn't matter: this message is intended in spirit
to be shown when creation succeeds, but with insane settings.

Signed-off-by: Glauber Costa <glommer@parallels.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: list_del_init() on removed events
Greg Thelen [Wed, 28 Nov 2012 21:50:45 +0000 (13:50 -0800)]
cgroup: list_del_init() on removed events

Use list_del_init() rather than list_del() to remove events from
cgrp->event_list.  No functional change.  This is just defensive
coding.

Signed-off-by: Greg Thelen <gthelen@google.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: fix lockdep warning for event_control
Greg Thelen [Wed, 28 Nov 2012 21:50:44 +0000 (13:50 -0800)]
cgroup: fix lockdep warning for event_control

The cgroup_event_wake() function is called with the wait queue head
locked and it takes cgrp->event_list_lock. However, in cgroup_rmdir()
remove_wait_queue() was being called after taking
cgrp->event_list_lock.  Correct the lock ordering by using a temporary
list to obtain the event list to remove from the wait queue.

Signed-off-by: Greg Thelen <gthelen@google.com>
Signed-off-by: Aaron Durbin <adurbin@google.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: move list add after list head initilization
Li Zhong [Wed, 28 Nov 2012 09:15:21 +0000 (17:15 +0800)]
cgroup: move list add after list head initilization

2243076ad1 ("cgroup: initialize cgrp->allcg_node in
init_cgroup_housekeeping()") initializes cgrp->allcg_node in
init_cgroup_housekeeping().  Then in init_cgroup_root(), we should
call init_cgroup_housekeeping() before adding it to &root->allcg_list;
otherwise, we are initializing an entry already in a list.

Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agonetprio_cgroup: allow nesting and inherit config on cgroup creation
Tejun Heo [Thu, 22 Nov 2012 15:32:47 +0000 (07:32 -0800)]
netprio_cgroup: allow nesting and inherit config on cgroup creation

Inherit netprio configuration from ->css_online(), allow nesting and
remove .broken_hierarchy marking.  This makes netprio_cgroup's
behavior match netcls_cgroup's.

Note that this patch changes userland-visible behavior.  Nesting is
allowed and the first level cgroups below the root cgroup behave
differently - they inherit priorities from the root cgroup on creation
instead of starting with 0.  This is unfortunate but not doing so is
much crazier.

Signed-off-by: Tejun Heo <tj@kernel.org>
Tested-and-Acked-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: David S. Miller <davem@davemloft.net>
11 years agonetprio_cgroup: implement netprio[_set]_prio() helpers
Tejun Heo [Thu, 22 Nov 2012 15:32:47 +0000 (07:32 -0800)]
netprio_cgroup: implement netprio[_set]_prio() helpers

Introduce two helpers - netprio_prio() and netprio_set_prio() - which
hide the details of priomap access and expansion.  This will help
implementing hierarchy support.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-and-Acked-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: David S. Miller <davem@davemloft.net>
11 years agonetprio_cgroup: use cgroup->id instead of cgroup_netprio_state->prioidx
Tejun Heo [Thu, 22 Nov 2012 15:32:47 +0000 (07:32 -0800)]
netprio_cgroup: use cgroup->id instead of cgroup_netprio_state->prioidx

With priomap expansion no longer depending on knowing max id
allocated, netprio_cgroup can use cgroup->id insted of cs->prioidx.
Drop prioidx alloc/free logic and convert all uses to cgroup->id.

* In cgrp_css_alloc(), parent->id test is moved above @cs allocation
  to simplify error path.

* In cgrp_css_free(), @cs assignment is made initialization.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-and-Acked-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: David S. Miller <davem@davemloft.net>
11 years agonetprio_cgroup: reimplement priomap expansion
Tejun Heo [Thu, 22 Nov 2012 15:32:46 +0000 (07:32 -0800)]
netprio_cgroup: reimplement priomap expansion

netprio kept track of the highest prioidx allocated and resized
priomaps accordingly when necessary.  This makes it necessary to keep
track of prioidx allocation and may end up resizing on every new
prioidx.

Update extend_netdev_table() such that it takes @target_idx which the
priomap should be able to accomodate.  If the priomap is large enough,
nothing happens; otherwise, the size is doubled until @target_idx can
be accomodated.

This makes max_prioidx and write_update_netdev_table() unnecessary.
write_priomap() now calls extend_netdev_table() directly.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-and-Acked-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: David S. Miller <davem@davemloft.net>
11 years agonetprio_cgroup: shorten variable names in extend_netdev_table()
Tejun Heo [Thu, 22 Nov 2012 15:32:46 +0000 (07:32 -0800)]
netprio_cgroup: shorten variable names in extend_netdev_table()

The function is about to go through a rewrite.  In preparation,
shorten the variable names so that we don't repeat "priomap" so often.

This patch is cosmetic.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-and-Acked-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: David S. Miller <davem@davemloft.net>
11 years agonetprio_cgroup: simplify write_priomap()
Tejun Heo [Thu, 22 Nov 2012 15:32:46 +0000 (07:32 -0800)]
netprio_cgroup: simplify write_priomap()

sscanf() doesn't bite.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-and-Acked-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: David S. Miller <davem@davemloft.net>
11 years agonetcls_cgroup: move config inheritance to ->css_online() and remove .broken_hierarchy...
Tejun Heo [Thu, 22 Nov 2012 15:32:46 +0000 (07:32 -0800)]
netcls_cgroup: move config inheritance to ->css_online() and remove .broken_hierarchy marking

It turns out that we'll have to live with attributes which are
inherited at cgroup creation time but not affected by further updates
to the parent afterwards - such attributes are already in wide use
e.g. for cpuset.

So, there's nothing to do for netcls_cgroup for hierarchy support.
Its current behavior - inherit only during creation - is good enough.

Move config inheriting from ->css_alloc() to ->css_online() for
consistency, which doesn't change behavior at all, and remove
.broken_hierarchy marking.

Signed-off-by: Tejun Heo <tj@kernel.org>
Tested-and-Acked-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: David S. Miller <davem@davemloft.net>
11 years agocgroup: remove obsolete guarantee from cgroup_task_migrate.
Tao Ma [Tue, 20 Nov 2012 14:06:18 +0000 (22:06 +0800)]
cgroup: remove obsolete guarantee from cgroup_task_migrate.

'guarantee' is already removed from cgroup_task_migrate, so remove
the corresponding comments. Some other typos in cgroup are also
changed.

Cc: Tejun Heo <tj@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: add cgroup->id
Tejun Heo [Mon, 19 Nov 2012 17:02:12 +0000 (09:02 -0800)]
cgroup: add cgroup->id

With the introduction of generic cgroup hierarchy iterators, css_id is
being phased out.  It was unnecessarily complex, id'ing the wrong
thing (cgroups need IDs, not CSSes) and has other oddities like not
being available at ->css_alloc().

This patch adds cgroup->id, which is a simple per-hierarchy
ida-allocated ID which is assigned before ->css_alloc() and released
after ->css_free().

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
11 years agocgroup, cpuset: remove cgroup_subsys->post_clone()
Tejun Heo [Mon, 19 Nov 2012 16:13:39 +0000 (08:13 -0800)]
cgroup, cpuset: remove cgroup_subsys->post_clone()

Currently CGRP_CPUSET_CLONE_CHILDREN triggers ->post_clone().  Now
that clone_children is cpuset specific, there's no reason to have this
rather odd option activation mechanism in cgroup core.  cpuset can
check the flag from its ->css_allocate() and take the necessary
action.

Move cpuset_post_clone() logic to the end of cpuset_css_alloc() and
remove cgroup_subsys->post_clone().

Loosely based on Glauber's "generalize post_clone into post_create"
patch.

Signed-off-by: Tejun Heo <tj@kernel.org>
Original-patch-by: Glauber Costa <glommer@parallels.com>
Original-patch: <1351686554-22592-2-git-send-email-glommer@parallels.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Glauber Costa <glommer@parallels.com>
11 years agocgroup: s/CGRP_CLONE_CHILDREN/CGRP_CPUSET_CLONE_CHILDREN/
Tejun Heo [Mon, 19 Nov 2012 16:13:38 +0000 (08:13 -0800)]
cgroup: s/CGRP_CLONE_CHILDREN/CGRP_CPUSET_CLONE_CHILDREN/

clone_children is only meaningful for cpuset and will stay that way.
Rename the flag to reflect that and update documentation.  Also, drop
clone_children() wrapper in cgroup.c.  The thin wrapper is used only a
few times and one of them will go away soon.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Glauber Costa <glommer@parallels.com>
11 years agocgroup: rename ->create/post_create/pre_destroy/destroy() to ->css_alloc/online/offli...
Tejun Heo [Mon, 19 Nov 2012 16:13:38 +0000 (08:13 -0800)]
cgroup: rename ->create/post_create/pre_destroy/destroy() to ->css_alloc/online/offline/free()

Rename cgroup_subsys css lifetime related callbacks to better describe
what their roles are.  Also, update documentation.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: allow ->post_create() to fail
Tejun Heo [Mon, 19 Nov 2012 16:13:38 +0000 (08:13 -0800)]
cgroup: allow ->post_create() to fail

There could be cases where controllers want to do initialization
operations which may fail from ->post_create().  This patch makes
->post_create() return -errno to indicate failure and online_css()
relay such failures.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Glauber Costa <glommer@parallels.com>
11 years agocgroup: update cgroup_create() failure path
Tejun Heo [Mon, 19 Nov 2012 16:13:38 +0000 (08:13 -0800)]
cgroup: update cgroup_create() failure path

cgroup_create() was ignoring failure of cgroupfs files.  Update it
such that, if file creation fails, it rolls back by calling
cgroup_destroy_locked() and returns failure.

Note that error out goto labels are renamed.  The labels are a bit
confusing but will become better w/ later cgroup operation renames.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: use mutex_trylock() when grabbing i_mutex of a new cgroup directory
Tejun Heo [Mon, 19 Nov 2012 16:13:37 +0000 (08:13 -0800)]
cgroup: use mutex_trylock() when grabbing i_mutex of a new cgroup directory

All cgroup directory i_mutexes nest outside cgroup_mutex; however, new
directory creation is a special case.  A new cgroup directory is
created while holding cgroup_mutex.  Populating the new directory
requires both the new directory's i_mutex and cgroup_mutex.  Because
all directory i_mutexes nest outside cgroup_mutex, grabbing both
requires releasing cgroup_mutex first, which isn't a good idea as the
new cgroup isn't yet ready to be manipulated by other cgroup
opreations.

This is worked around by grabbing the new directory's i_mutex while
holding cgroup_mutex before making it visible.  As there's no other
user at that point, grabbing the i_mutex under cgroup_mutex can't lead
to deadlock.

cgroup_create_file() was using I_MUTEX_CHILD to tell lockdep not to
worry about the reverse locking order; however, this creates pseudo
locking dependency cgroup_mutex -> I_MUTEX_CHILD, which isn't true -
all directory i_mutexes are still nested outside cgroup_mutex.  This
pseudo locking dependency can lead to spurious lockdep warnings.

Use mutex_trylock() instead.  This will always succeed and lockdep
doesn't create any locking dependency for it.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: simplify cgroup_load_subsys() failure path
Tejun Heo [Mon, 19 Nov 2012 16:13:37 +0000 (08:13 -0800)]
cgroup: simplify cgroup_load_subsys() failure path

Now that cgroup_unload_subsys() can tell whether the root css is
online or not, we can safely call cgroup_unload_subsys() after idr
init failure in cgroup_load_subsys().

Replace the manual unrolling and invoke cgroup_unload_subsys() on
failure.  This drops cgroup_mutex inbetween but should be safe as the
subsystem will fail try_module_get() and thus can't be mounted
inbetween.  As this means that cgroup_unload_subsys() can be called
before css_sets are rehashed, remove BUG_ON() on %NULL
css_set->subsys[] from cgroup_unload_subsys().

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: introduce CSS_ONLINE flag and on/offline_css() helpers
Tejun Heo [Mon, 19 Nov 2012 16:13:37 +0000 (08:13 -0800)]
cgroup: introduce CSS_ONLINE flag and on/offline_css() helpers

New helpers on/offline_css() respectively wrap ->post_create() and
->pre_destroy() invocations.  online_css() sets CSS_ONLINE after
->post_create() is complete and offline_css() invokes ->pre_destroy()
iff CSS_ONLINE is set and clears it while also handling the temporary
dropping of cgroup_mutex.

This patch doesn't introduce any behavior change at the moment but
will be used to improve cgroup_create() failure path and allow
->post_create() to fail.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: separate out cgroup_destroy_locked()
Tejun Heo [Mon, 19 Nov 2012 16:13:37 +0000 (08:13 -0800)]
cgroup: separate out cgroup_destroy_locked()

Separate out cgroup_destroy_locked() from cgroup_destroy().  This will
be later used in cgroup_create() failure path.

While at it, add lockdep asserts on i_mutex and cgroup_mutex, and move
@d and @parent assignments to their declarations.

This patch doesn't introduce any functional difference.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: fix harmless bugs in cgroup_load_subsys() fail path and cgroup_unload_subsys()
Tejun Heo [Mon, 19 Nov 2012 16:13:37 +0000 (08:13 -0800)]
cgroup: fix harmless bugs in cgroup_load_subsys() fail path and cgroup_unload_subsys()

* If idr init fails, cgroup_load_subsys() cleared dummytop->subsys[]
  before calilng ->destroy() making CSS inaccessible to the callback,
  and didn't unlink ss->sibling.  As no modular controller uses
  ->use_id, this doesn't cause any actual problems.

* cgroup_unload_subsys() was forgetting to free idr, call
  ->pre_destroy() and clear ->active.  As there currently is no
  modular controller which uses ->use_id, ->pre_destroy() or ->active,
  this doesn't cause any actual problems.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: lock cgroup_mutex in cgroup_init_subsys()
Tejun Heo [Mon, 19 Nov 2012 16:13:36 +0000 (08:13 -0800)]
cgroup: lock cgroup_mutex in cgroup_init_subsys()

Make cgroup_init_subsys() grab cgroup_mutex while initializing a
subsystem so that all helpers and callbacks are called under the
context they expect.  This isn't strictly necessary as
cgroup_init_subsys() doesn't race with anybody but will allow adding
lockdep assertions.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: trivial cleanup for cgroup_init/load_subsys()
Tejun Heo [Mon, 19 Nov 2012 16:13:36 +0000 (08:13 -0800)]
cgroup: trivial cleanup for cgroup_init/load_subsys()

Consistently use @css and @dummytop in these two functions instead of
referring to them indirectly.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: make CSS_* flags bit masks instead of bit positions
Tejun Heo [Mon, 19 Nov 2012 16:13:36 +0000 (08:13 -0800)]
cgroup: make CSS_* flags bit masks instead of bit positions

Currently, CSS_* flags are defined as bit positions and manipulated
using atomic bitops.  There's no reason to use atomic bitops for them
and bit positions are clunkier to deal with than bit masks.  Make
CSS_* bit masks instead and use the usual C bitwise operators to
access them.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: cgroup->dentry isn't a RCU pointer
Tejun Heo [Mon, 19 Nov 2012 16:13:36 +0000 (08:13 -0800)]
cgroup: cgroup->dentry isn't a RCU pointer

cgroup->dentry is marked and used as a RCU pointer; however, it isn't
one - the final dentry put doesn't go through call_rcu().  cgroup and
dentry share the same RCU freeing rule via synchronize_rcu() in
cgroup_diput() (kfree_rcu() used on cgrp is unnecessary).  If cgrp is
accessible under RCU read lock, so is its dentry and dereferencing
cgrp->dentry doesn't need any further RCU protection or annotation.

While not being accurate, before the previous patch, the RCU accessors
served a purpose as memory barriers - cgroup->dentry used to be
assigned after the cgroup was made visible to cgroup_path(), so the
assignment and dereferencing in cgroup_path() needed the memory
barrier pair.  Now that list_add_tail_rcu() happens after
cgroup->dentry is assigned, this no longer is necessary.

Remove the now unnecessary and misleading RCU annotations from
cgroup->dentry.  To make up for the removal of rcu_dereference_check()
in cgroup_path(), add an explicit rcu_lockdep_assert(), which asserts
the dereference rule of @cgrp, not cgrp->dentry.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: create directory before linking while creating a new cgroup
Tejun Heo [Mon, 19 Nov 2012 16:13:36 +0000 (08:13 -0800)]
cgroup: create directory before linking while creating a new cgroup

While creating a new cgroup, cgroup_create() links the newly allocated
cgroup into various places before trying to create its directory.
Because cgroup life-cycle is tied to the vfs objects, this makes it
impossible to use cgroup_rmdir() for rolling back creation - the
removal logic depends on having full vfs objects.

This patch moves directory creation above linking and collect linking
operations to one place.  This allows directory creation failure to
share error exit path with css allocation failures and any failure
sites afterwards (to be added later) can use cgroup_rmdir() logic to
undo creation.

Note that this also makes the memory barriers around cgroup->dentry,
which currently is misleadingly using RCU operations, unnecessary.
This will be handled in the next patch.

While at it, locking BUG_ON() on i_mutex is converted to
lockdep_assert_held().

v2: Patch originally removed %NULL dentry check in cgroup_path();
    however, Li pointed out that this patch doesn't make it
    unnecessary as ->create() may call cgroup_path().  Drop the
    change for now.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: open-code cgroup_create_dir()
Tejun Heo [Mon, 19 Nov 2012 16:13:36 +0000 (08:13 -0800)]
cgroup: open-code cgroup_create_dir()

The operation order of cgroup creation is about to change and
cgroup_create_dir() is more of a hindrance than a proper abstraction.
Open-code it by moving the parent nlink adjustment next to self nlink
adjustment in cgroup_create_file() and the rest to cgroup_create().

This patch doesn't introduce any behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: initialize cgrp->allcg_node in init_cgroup_housekeeping()
Tejun Heo [Mon, 19 Nov 2012 16:13:35 +0000 (08:13 -0800)]
cgroup: initialize cgrp->allcg_node in init_cgroup_housekeeping()

Not strictly necessary but it's annoying to have uninitialized
list_head around.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: remove incorrect dget/dput() pair in cgroup_create_dir()
Tejun Heo [Mon, 19 Nov 2012 16:13:35 +0000 (08:13 -0800)]
cgroup: remove incorrect dget/dput() pair in cgroup_create_dir()

cgroup_create_dir() does weird dancing with dentry refcnt.  On
success, it gets and then puts it achieving nothing.  On failure, it
puts but there isn't no matching get anywhere leading to the following
oops if cgroup_create_file() fails for whatever reason.

  ------------[ cut here ]------------
  kernel BUG at /work/os/work/fs/dcache.c:552!
  invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
  Modules linked in:
  CPU 2
  Pid: 697, comm: mkdir Not tainted 3.7.0-rc4-work+ #3 Bochs Bochs
  RIP: 0010:[<ffffffff811d9c0c>]  [<ffffffff811d9c0c>] dput+0x1dc/0x1e0
  RSP: 0018:ffff88001a3ebef8  EFLAGS: 00010246
  RAX: 0000000000000000 RBX: ffff88000e5b1ef8 RCX: 0000000000000403
  RDX: 0000000000000303 RSI: 2000000000000000 RDI: ffff88000e5b1f58
  RBP: ffff88001a3ebf18 R08: ffffffff82c76960 R09: 0000000000000001
  R10: ffff880015022080 R11: ffd9bed70f48a041 R12: 00000000ffffffea
  R13: 0000000000000001 R14: ffff88000e5b1f58 R15: 00007fff57656d60
  FS:  00007ff05fcb3800(0000) GS:ffff88001fd00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00000000004046f0 CR3: 000000001315f000 CR4: 00000000000006e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
  Process mkdir (pid: 697, threadinfo ffff88001a3ea000, task ffff880015022080)
  Stack:
   ffff88001a3ebf48 00000000ffffffea 0000000000000001 0000000000000000
   ffff88001a3ebf38 ffffffff811cc889 0000000000000001 ffff88000e5b1ef8
   ffff88001a3ebf68 ffffffff811d1fc9 ffff8800198d7f18 ffff880019106ef8
  Call Trace:
   [<ffffffff811cc889>] done_path_create+0x19/0x50
   [<ffffffff811d1fc9>] sys_mkdirat+0x59/0x80
   [<ffffffff811d2009>] sys_mkdir+0x19/0x20
   [<ffffffff81be1e02>] system_call_fastpath+0x16/0x1b
  Code: 00 48 8d 90 18 01 00 00 48 89 93 c0 00 00 00 4c 89 a0 18 01 00 00 48 8b 83 a0 00 00 00 83 80 28 01 00 00 01 e8 e6 6f a0 00 eb 92 <0f> 0b 66 90 0f 1f 44 00 00 55 48 89 e5 41 57 41 56 49 89 fe 41
  RIP  [<ffffffff811d9c0c>] dput+0x1dc/0x1e0
   RSP <ffff88001a3ebef8>
  ---[ end trace 1277bcfd9561ddb0 ]---

Fix it by dropping the unnecessary dget/dput() pair.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: stable@vger.kernel.org
12 years agocgroup_freezer: implement proper hierarchy support
Tejun Heo [Fri, 9 Nov 2012 17:12:30 +0000 (09:12 -0800)]
cgroup_freezer: implement proper hierarchy support

Up until now, cgroup_freezer didn't implement hierarchy properly.
cgroups could be arranged in hierarchy but it didn't make any
difference in how each cgroup_freezer behaved.  They all operated
separately.

This patch implements proper hierarchy support.  If a cgroup is
frozen, all its descendants are frozen.  A cgroup is thawed iff it and
all its ancestors are THAWED.  freezer.self_freezing shows the current
freezing state for the cgroup itself.  freezer.parent_freezing shows
whether the cgroup is freezing because any of its ancestors is
freezing.

freezer_post_create() locks the parent and new cgroup and inherits the
parent's state and freezer_change_state() applies new state top-down
using cgroup_for_each_descendant_pre() which guarantees that no child
can escape its parent's state.  update_if_frozen() uses
cgroup_for_each_descendant_post() to propagate frozen states
bottom-up.

Synchronization could be coarser and easier by using a single mutex to
protect all hierarchy operations.  Finer grained approach was used
because it wasn't too difficult for cgroup_freezer and I think it's
beneficial to have an example implementation and cgroup_freezer is
rather simple and can serve a good one.

As this makes cgroup_freezer properly hierarchical,
freezer_subsys.broken_hierarchy marking is removed.

Note that this patch changes userland visible behavior - freezing a
cgroup now freezes all its descendants too.  This behavior change is
intended and has been warned via .broken_hierarchy.

v2: Michal spotted a bug in freezer_change_state() - descendants were
    inheriting from the wrong ancestor.  Fixed.

v3: Documentation/cgroups/freezer-subsystem.txt updated.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
12 years agocgroup_freezer: add ->post_create() and ->pre_destroy() and track online state
Tejun Heo [Fri, 9 Nov 2012 17:12:30 +0000 (09:12 -0800)]
cgroup_freezer: add ->post_create() and ->pre_destroy() and track online state

A cgroup is online and visible to iteration between ->post_create()
and ->pre_destroy().  This patch introduces CGROUP_FREEZER_ONLINE and
toggles it from the newly added freezer_post_create() and
freezer_pre_destroy() while holding freezer->lock such that a
cgroup_freezer can be reilably distinguished to be online.  This will
be used by full hierarchy support.

ONLINE test is added to freezer_apply_state() but it currently doesn't
make any difference as freezer_write() can only be called for an
online cgroup.

Adjusting system_freezing_cnt on destruction is moved from
freezer_destroy() to the new freezer_pre_destroy() for consistency.

This patch doesn't introduce any noticeable behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
12 years agocgroup_freezer: introduce CGROUP_FREEZING_[SELF|PARENT]
Tejun Heo [Fri, 9 Nov 2012 17:12:30 +0000 (09:12 -0800)]
cgroup_freezer: introduce CGROUP_FREEZING_[SELF|PARENT]

Introduce FREEZING_SELF and FREEZING_PARENT and make FREEZING OR of
the two flags.  This is to prepare for full hierarchy support.

freezer_apply_date() is updated such that it can handle setting and
clearing of both flags.  The two flags are also exposed to userland
via read-only files self_freezing and parent_freezing.

Other than the added cgroupfs files, this patch doesn't introduce any
behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
12 years agocgroup_freezer: make freezer->state mask of flags
Tejun Heo [Fri, 9 Nov 2012 17:12:30 +0000 (09:12 -0800)]
cgroup_freezer: make freezer->state mask of flags

freezer->state was an enum value - one of THAWED, FREEZING and FROZEN.
As the scheduled full hierarchy support requires more than one
freezing condition, switch it to mask of flags.  If FREEZING is not
set, it's thawed.  FREEZING is set if freezing or frozen.  If frozen,
both FREEZING and FROZEN are set.  Now that tasks can be attached to
an already frozen cgroup, this also makes freezing condition checks
more natural.

This patch doesn't introduce any behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
12 years agocgroup_freezer: prepare freezer_change_state() for full hierarchy support
Tejun Heo [Fri, 9 Nov 2012 17:12:30 +0000 (09:12 -0800)]
cgroup_freezer: prepare freezer_change_state() for full hierarchy support

* Make freezer_change_state() take bool @freeze instead of enum
  freezer_state.

* Separate out freezer_apply_state() out of freezer_change_state().
  This makes freezer_change_state() a rather silly thin wrapper.  It
  will be filled with hierarchy handling later on.

This patch doesn't introduce any behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
12 years agocgroup_freezer: trivial cleanups
Tejun Heo [Fri, 9 Nov 2012 17:12:29 +0000 (09:12 -0800)]
cgroup_freezer: trivial cleanups

* Clean-up indentation and line-breaks.  Drop the invalid comment
  about freezer->lock.

* Make all internal functions take @freezer instead of both @cgroup
  and @freezer.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
12 years agocgroup: implement generic child / descendant walk macros
Tejun Heo [Fri, 9 Nov 2012 17:12:29 +0000 (09:12 -0800)]
cgroup: implement generic child / descendant walk macros

Currently, cgroup doesn't provide any generic helper for walking a
given cgroup's children or descendants.  This patch adds the following
three macros.

* cgroup_for_each_child() - walk immediate children of a cgroup.

* cgroup_for_each_descendant_pre() - visit all descendants of a cgroup
  in pre-order tree traversal.

* cgroup_for_each_descendant_post() - visit all descendants of a
  cgroup in post-order tree traversal.

All three only require the user to hold RCU read lock during
traversal.  Verifying that each iterated cgroup is online is the
responsibility of the user.  When used with proper synchronization,
cgroup_for_each_descendant_pre() can be used to propagate state
updates to descendants in reliable way.  See comments for details.

v2: s/config/state/ in commit message and comments per Michal.  More
    documentation on synchronization rules.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujisu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Li Zefan <lizefan@huawei.com>
12 years agocgroup: use rculist ops for cgroup->children
Tejun Heo [Fri, 9 Nov 2012 17:12:29 +0000 (09:12 -0800)]
cgroup: use rculist ops for cgroup->children

Use RCU safe list operations for cgroup->children.  This will be used
to implement cgroup children / descendant walking which can be used by
controllers.

Note that cgroup_create() now puts a new cgroup at the end of the
->children list instead of head.  This isn't strictly necessary but is
done so that the iteration order is more conventional.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
12 years agocgroup: add cgroup_subsys->post_create()
Tejun Heo [Fri, 9 Nov 2012 17:12:29 +0000 (09:12 -0800)]
cgroup: add cgroup_subsys->post_create()

Currently, there's no way for a controller to find out whether a new
cgroup finished all ->create() allocatinos successfully and is
considered "live" by cgroup.

This becomes a problem later when we add generic descendants walking
to cgroup which can be used by controllers as controllers don't have a
synchronization point where it can synchronize against new cgroups
appearing in such walks.

This patch adds ->post_create().  It's called after all ->create()
succeeded and the cgroup is linked into the generic cgroup hierarchy.
This plays the counterpart of ->pre_destroy().

When used in combination with the to-be-added generic descendant
iterators, ->post_create() can be used to implement reliable state
inheritance.  It will be explained with the descendant iterators.

v2: Added a paragraph about its future use w/ descendant iterators per
    Michal.

v3: Forgot to add ->post_create() invocation to cgroup_load_subsys().
    Fixed.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Glauber Costa <glommer@parallels.com>
12 years agocgroup: set 'start' with the right value in cgroup_path.
Tao Ma [Thu, 8 Nov 2012 13:36:38 +0000 (21:36 +0800)]
cgroup: set 'start' with the right value in cgroup_path.

'start' is set to buf + buflen and do the '--' immediately.
Just set it to 'buf + buflen - 1' directly.

Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
12 years agodevice_cgroup: add lockdep asserts
Tejun Heo [Tue, 6 Nov 2012 17:16:53 +0000 (09:16 -0800)]
device_cgroup: add lockdep asserts

device_cgroup uses RCU safe ->exceptions list which is write-protected
by devcgroup_mutex and has had some issues using locking correctly.
Add lockdep asserts to utility functions so that future errors can be
easily detected.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Cc: Aristeu Rozanski <aris@redhat.com>
Cc: Li Zefan <lizefan@huawei.com>
12 years agoMerge branch 'cgroup/for-3.7-fixes' into cgroup/for-3.8
Tejun Heo [Tue, 6 Nov 2012 20:26:23 +0000 (12:26 -0800)]
Merge branch 'cgroup/for-3.7-fixes' into cgroup/for-3.8

This is to receive device_cgroup fixes so that further device_cgroup
changes can be made in cgroup/for-3.8.

Signed-off-by: Tejun Heo <tj@kernel.org>
12 years agodevice_cgroup: fix RCU usage
Tejun Heo [Tue, 6 Nov 2012 17:17:37 +0000 (09:17 -0800)]
device_cgroup: fix RCU usage

dev_cgroup->exceptions is protected with devcgroup_mutex for writes
and RCU for reads; however, RCU usage isn't correct.

* dev_exception_clean() doesn't use RCU variant of list_del() and
  kfree().  The function can race with may_access() and may_access()
  may end up dereferencing already freed memory.  Use list_del_rcu()
  and kfree_rcu() instead.

* may_access() may be called only with RCU read locked but doesn't use
  RCU safe traversal over ->exceptions.  Use list_for_each_entry_rcu().

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Cc: stable@vger.kernel.org
Cc: Aristeu Rozanski <aris@redhat.com>
Cc: Li Zefan <lizefan@huawei.com>
12 years agodevice_cgroup: fix unchecked cgroup parent usage
Aristeu Rozanski [Tue, 6 Nov 2012 15:25:04 +0000 (07:25 -0800)]
device_cgroup: fix unchecked cgroup parent usage

In 4cef7299b478687 ("device_cgroup: add proper checking when changing
default behavior") the cgroup parent usage is unchecked.  root will not
have a parent and trying to use device.{allow,deny} will cause problems.
For some reason my stressing scripts didn't test the root directory so I
didn't catch it on my regular tests.

Signed-off-by: Aristeu Rozanski <aris@redhat.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: James Morris <jmorris@namei.org>
Cc: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
12 years agoMerge branch 'cgroup-rmdir-updates' into cgroup/for-3.8
Tejun Heo [Mon, 5 Nov 2012 17:21:51 +0000 (09:21 -0800)]
Merge branch 'cgroup-rmdir-updates' into cgroup/for-3.8

Pull rmdir updates into for-3.8 so that further callback updates can
be put on top.  This pull created a trivial conflict between the
following two commits.

  8c7f6edbda ("cgroup: mark subsystems with broken hierarchy support and whine if cgroups are nested for them")
  ed95779340 ("cgroup: kill cgroup_subsys->__DEPRECATED_clear_css_refs")

The former added a field to cgroup_subsys and the latter removed one
from it.  They happen to be colocated causing the conflict.  Keeping
what's added and removing what's removed resolves the conflict.

Signed-off-by: Tejun Heo <tj@kernel.org>
12 years agocgroup: make ->pre_destroy() return void
Tejun Heo [Mon, 5 Nov 2012 17:16:59 +0000 (09:16 -0800)]
cgroup: make ->pre_destroy() return void

All ->pre_destory() implementations return 0 now, which is the only
allowed return value.  Make it return void.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
12 years agohugetlb: do not fail in hugetlb_cgroup_pre_destroy
Michal Hocko [Fri, 26 Oct 2012 11:37:33 +0000 (13:37 +0200)]
hugetlb: do not fail in hugetlb_cgroup_pre_destroy

Now that pre_destroy callbacks are called from the context where neither
any task can attach the group nor any children group can be added there
is no other way to fail from hugetlb_pre_destroy.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Glauber Costa <glommer@parallels.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
12 years agomemcg: make mem_cgroup_reparent_charges non failing
Michal Hocko [Fri, 26 Oct 2012 11:37:32 +0000 (13:37 +0200)]
memcg: make mem_cgroup_reparent_charges non failing

Now that pre_destroy callbacks are called from the context where neither
any task can attach the group nor any children group can be added there
is no other way to fail from mem_cgroup_pre_destroy.
mem_cgroup_pre_destroy doesn't have to take a reference to memcg's css
because all css' are marked dead already.

tj: Remove now unused local variable @cgrp from
    mem_cgroup_reparent_charges().

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Glauber Costa <glommer@parallels.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
12 years agocgroup: remove CGRP_WAIT_ON_RMDIR, cgroup_exclude_rmdir() and cgroup_release_and_wake...
Tejun Heo [Mon, 5 Nov 2012 17:16:59 +0000 (09:16 -0800)]
cgroup: remove CGRP_WAIT_ON_RMDIR, cgroup_exclude_rmdir() and cgroup_release_and_wakeup_rmdir()

CGRP_WAIT_ON_RMDIR is another kludge which was added to make cgroup
destruction rollback somewhat working.  cgroup_rmdir() used to drain
CSS references and CGRP_WAIT_ON_RMDIR and the associated waitqueue and
helpers were used to allow the task performing rmdir to wait for the
next relevant event.

Unfortunately, the wait is visible to controllers too and the
mechanism got exposed to memcg by 887032670d ("cgroup avoid permanent
sleep at rmdir").

Now that the draining and retries are gone, CGRP_WAIT_ON_RMDIR is
unnecessary.  Remove it and all the mechanisms supporting it.  Note
that memcontrol.c changes are essentially revert of 887032670d
("cgroup avoid permanent sleep at rmdir").

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Balbir Singh <bsingharora@gmail.com>
12 years agocgroup: deactivate CSS's and mark cgroup dead before invoking ->pre_destroy()
Tejun Heo [Mon, 5 Nov 2012 17:16:59 +0000 (09:16 -0800)]
cgroup: deactivate CSS's and mark cgroup dead before invoking ->pre_destroy()

Because ->pre_destroy() could fail and can't be called under
cgroup_mutex, cgroup destruction did something very ugly.

  1. Grab cgroup_mutex and verify it can be destroyed; fail otherwise.

  2. Release cgroup_mutex and call ->pre_destroy().

  3. Re-grab cgroup_mutex and verify it can still be destroyed; fail
     otherwise.

  4. Continue destroying.

In addition to being ugly, it has been always broken in various ways.
For example, memcg ->pre_destroy() expects the cgroup to be inactive
after it's done but tasks can be attached and detached between #2 and
#3 and the conditions that memcg verified in ->pre_destroy() might no
longer hold by the time control reaches #3.

Now that ->pre_destroy() is no longer allowed to fail.  We can switch
to the following.

  1. Grab cgroup_mutex and verify it can be destroyed; fail otherwise.

  2. Deactivate CSS's and mark the cgroup removed thus preventing any
     further operations which can invalidate the verification from #1.

  3. Release cgroup_mutex and call ->pre_destroy().

  4. Re-grab cgroup_mutex and continue destroying.

After this change, controllers can safely assume that ->pre_destroy()
will only be called only once for a given cgroup and, once
->pre_destroy() is called, the cgroup will stay dormant till it's
destroyed.

This removes the only reason ->pre_destroy() can fail - new task being
attached or child cgroup being created inbetween.  Error out path is
removed and ->pre_destroy() invocation is open coded in
cgroup_rmdir().

v2: cgroup_call_pre_destroy() removal moved to this patch per Michal.
    Commit message updated per Glauber.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Glauber Costa <glommer@parallels.com>
12 years agocgroup: use cgroup_lock_live_group(parent) in cgroup_create()
Tejun Heo [Mon, 5 Nov 2012 17:16:59 +0000 (09:16 -0800)]
cgroup: use cgroup_lock_live_group(parent) in cgroup_create()

This patch makes cgroup_create() fail if @parent is marked removed.
This is to prepare for further updates to cgroup_rmdir() path.

Note that this change isn't strictly necessary.  cgroup can only be
created via mkdir and the removed marking and dentry removal happen
without releasing cgroup_mutex, so cgroup_create() can never race with
cgroup_rmdir().  Even after the scheduled updates to cgroup_rmdir(),
cgroup_mkdir() and cgroup_rmdir() are synchronized by i_mutex
rendering the added liveliness check unnecessary.

Do it anyway such that locking is contained inside cgroup proper and
we don't get nasty surprises if we ever grow another caller of
cgroup_create().

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
12 years agocgroup: kill CSS_REMOVED
Tejun Heo [Mon, 5 Nov 2012 17:16:58 +0000 (09:16 -0800)]
cgroup: kill CSS_REMOVED

CSS_REMOVED is one of the several contortions which were necessary to
support css reference draining on cgroup removal.  All css->refcnts
which need draining should be deactivated and verified to equal zero
atomically w.r.t. css_tryget().  If any one isn't zero, all refcnts
needed to be re-activated and css_tryget() shouldn't fail in the
process.

This was achieved by letting css_tryget() busy-loop until either the
refcnt is reactivated (failed removal attempt) or CSS_REMOVED is set
(committing to removal).

Now that css refcnt draining is no longer used, there's no need for
atomic rollback mechanism.  css_tryget() simply can look at the
reference count and fail if it's deactivated - it's never getting
re-activated.

This patch removes CSS_REMOVED and updates __css_tryget() to fail if
the refcnt is deactivated.  As deactivation and removal are a single
step now, they no longer need to be protected against css_tryget()
happening from irq context.  Remove local_irq_disable/enable() from
cgroup_rmdir().

Note that this removes css_is_removed() whose only user is VM_BUG_ON()
in memcontrol.c.  We can replace it with a check on the refcnt but
given that the only use case is a debug assert, I think it's better to
simply unexport it.

v2: Comment updated and explanation on local_irq_disable/enable()
    added per Michal Hocko.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Balbir Singh <bsingharora@gmail.com>
12 years agocgroup: kill cgroup_subsys->__DEPRECATED_clear_css_refs
Tejun Heo [Mon, 5 Nov 2012 17:16:58 +0000 (09:16 -0800)]
cgroup: kill cgroup_subsys->__DEPRECATED_clear_css_refs

2ef37d3fe4 ("memcg: Simplify mem_cgroup_force_empty_list error
handling") removed the last user of __DEPRECATED_clear_css_refs.  This
patch removes __DEPRECATED_clear_css_refs and mechanisms to support
it.

* Conditionals dependent on __DEPRECATED_clear_css_refs removed.

* cgroup_clear_css_refs() can no longer fail.  All that needs to be
  done are deactivating refcnts, setting CSS_REMOVED and putting the
  base reference on each css.  Remove cgroup_clear_css_refs() and the
  failure path, and open-code the loops into cgroup_rmdir().

This patch keeps the two for_each_subsys() loops separate while open
coding them.  They can be merged now but there are scheduled changes
which need them to be separate, so keep them separate to reduce the
amount of churn.

local_irq_save/restore() from cgroup_clear_css_refs() are replaced
with local_irq_disable/enable() for simplicity.  This is safe as
cgroup_rmdir() is always called with IRQ enabled.  Note that this IRQ
switching is necessary to ensure that css_tryget() isn't called from
IRQ context on the same CPU while lower context is between CSS
deactivation and setting CSS_REMOVED as css_tryget() would hang
forever in such cases waiting for CSS to be re-activated or
CSS_REMOVED set.  This will go away soon.

v2: cgroup_call_pre_destroy() removal dropped per Michal.  Commit
    message updated to explain local_irq_disable/enable() conversion.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
12 years agoLinux 3.7-rc4
Linus Torvalds [Sun, 4 Nov 2012 19:07:39 +0000 (11:07 -0800)]
Linux 3.7-rc4

12 years agoMerge tag 'nfs-for-3.7-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Sat, 3 Nov 2012 22:27:21 +0000 (15:27 -0700)]
Merge tag 'nfs-for-3.7-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client bugfixes from Trond Myklebust:

 - Fix a bunch of deadlock situations:
   * State recovery can deadlock if we fail to release sequence ids
     before scheduling the recovery thread.
   * Calling deactivate_super() from an RPC workqueue thread can
     deadlock because of the call to rpc_shutdown_client.

 - Display the device name correctly in /proc/*/mounts

 - Fix a number of incorrect error return values:
   * When NFSv3 mounts fail due to a timeout.
   * On NFSv4.1 backchannel setup failure
   * On NFSv4 open access checks

 - pnfs_find_alloc_layout() must check the layout pointer for NULL

 - Fix a regression in the legacy DNS resolved

* tag 'nfs-for-3.7-4' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFS4: nfs4_opendata_access should return errno
  NFSv4: Initialise the NFSv4.1 slot table highest_used_slotid correctly
  SUNRPC: return proper errno from backchannel_rqst
  NFS: add nfs_sb_deactive_async to avoid deadlock
  nfs: Show original device name verbatim in /proc/*/mount{s,info}
  nfsv3: Make v3 mounts fail with ETIMEDOUTs instead EIO on mountd timeouts
  nfs: Check whether a layout pointer is NULL before free it
  NFS: fix bug in legacy DNS resolver.
  NFSv4: nfs4_locku_done must release the sequence id
  NFSv4.1: We must release the sequence id when we fail to get a session slot
  NFS: Wait for session recovery to finish before returning

12 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux
Linus Torvalds [Sat, 3 Nov 2012 22:25:14 +0000 (15:25 -0700)]
Merge branch 'release' of git://git./linux/kernel/git/rzhang/linux

Pull thermal management & ACPI update from Zhang Rui,

Ho humm.  Normally these things go through Len.  But it's just three
small fixes, I guess I can pull directly too.

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux:
  exynos4_tmu_driver_ids should be exynos_tmu_driver_ids.
  ACPI video: Ignore errors after _DOD evaluation.
  thermal: solve compilation errors in rcar_thermal

12 years agoMerge branch 'i2c-embedded/for-current' of git://git.pengutronix.de/git/wsa/linux
Linus Torvalds [Sat, 3 Nov 2012 22:14:54 +0000 (15:14 -0700)]
Merge branch 'i2c-embedded/for-current' of git://git.pengutronix.de/git/wsa/linux

Pull i2c embedded fixes from Wolfram Sang:
 "Two patches are usual stuff.

  The bigger patch is needed to correct a wrong decision made in this
  merge window.  We hoped to get the PIOQUEUE mode in the mxs driver
  working with DMA, but it turned out to be too broken (leading to data
  loss), so we now think it is best to remove it entirely and work only
  with DMA now.  The patch should be in 3.7.  IMO, so users never get
  the chance to use both modes in parallel."

* 'i2c-embedded/for-current' of git://git.pengutronix.de/git/wsa/linux:
  i2c: tegra: set irq name as device name
  i2c-nomadik: Fixup clock handling
  i2c: mxs: remove broken PIOQUEUE support

12 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Sat, 3 Nov 2012 22:13:49 +0000 (15:13 -0700)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "Scattered selection of fixes:

   - radeon: load detect fixes from SuSE/AMD
   - intel: misc i830, sdvo regression, vesafb kickoff ums fix
   - exynos: maintainers entry update + fixes
   - udl: fix stride scanout issue

  it's slightly bigger than I'd probably like, but nothing looked
  dangerous enough to hold off on."

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/udl: fix stride issues scanning out stride != width*bpp
  drm/radeon: add load detection support for ext DAC on R200 (v2)
  DRM/radeon: For single CRTC GPUs move handling of CRTC_CRT_ON to crtc_dpms().
  DRM/Radeon: Fix TV DAC Load Detection for single CRTC chips.
  DRM/Radeon: Clean up code in TV DAC load detection.
  drm/radeon: fix ATPX function documentation
  drivers/gpu/drm/radeon/evergreen_cs.c: Remove unnecessary semicolon
  DRM/Radeon: On DVI-I use Load Detection when EDID is bogus.
  DRM/Radeon: Fix primary DAC Load Detection for RV100 chips.
  DRM/Radeon: Fix Load Detection on legacy primary DAC.
  drm: exynos: removed warning due to missing typecast for mixer driver data
  drm/exynos: add support for ARCH_MULTIPLATFORM
  MAINTAINERS: Add git repository for Exynos DRM
  drm/exynos: fix display on issue
  drm/i915: Only kick out vesafb if we takeover the fbcon with KMS
  drm/i915: be less verbose about inability to provide vendor backlight
  drm/i915: clear the entire sdvo infoframe buffer
  drm/i915: VGA needs to be on pipe A on i830M
  drm/i915: fix overlay on i830M

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Sat, 3 Nov 2012 03:48:41 +0000 (20:48 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:
 "First post-Sandy pull request"

 1) Fix antenna gain handling and initialization of chan->max_reg_power
    in wireless, from Felix Fietkau.

 2) Fix nexthop handling in H.232 conntrack helper, from Julian
    Anastasov.

 3) Only process 80211 mesh config header in certain kinds of frames,
    from Javier Cardona.

 4) 80211 management frame header length needs to be validated, from
    Johannes Berg.

 5) Don't access free'd SKBs in ath9k driver, from Felix Fietkay.

 6) Test for permanent state correctly in VXLAN driver, from Stephen
    Hemminger.

 7) BNX2X bug fixes from Yaniv Rosner and Dmitry Kravkov.

 8) Fix off by one errors in bonding, from Nikolay ALeksandrov.

 9) Fix divide by zero in TCP-Illinois congestion control.  From Jesper
    Dangaard Brouer.

10) TCP metrics code says "Yo dawg, I heard you like sizeof, so I did a
    sizeof of a sizeof, so you can size your size" Fix from Julian
    Anastasov.

11) Several drivers do mdiobus_free without first doing an
    mdiobus_unregister leading to stray pointer references.  Fix from
    Peter Senna Tschudin.

12) Fix OOPS in l2tp_eth_create() error path, it's another danling
    pointer kinda situation.  Fix from Tom Parkin.

13) Hardware driven by the vmxnet driver can't handle larger than 16K
    fragments, so split them up when necessary.  From Eric Dumazet.

14) Handle zero length data length in tcp_send_rcvq() properly.  Fix
    from Pavel Emelyanov.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (38 commits)
  tcp-repair: Handle zero-length data put in rcv queue
  vmxnet3: must split too big fragments
  l2tp: fix oops in l2tp_eth_create() error path
  cxgb4: Fix unable to get UP event from the LLD
  drivers/net/phy/mdio-bitbang.c: Call mdiobus_unregister before mdiobus_free
  drivers/net/ethernet/nxp/lpc_eth.c: Call mdiobus_unregister before mdiobus_free
  bnx2x: fix HW initialization using fw 7.8.x
  tcp: Fix double sizeof in new tcp_metrics code
  net: fix divide by zero in tcp algorithm illinois
  net: sctp: Fix typo in net/sctp
  bonding: fix second off-by-one error
  bonding: fix off-by-one error
  bnx2x: Disable FCoE for 57840 since not yet supported by FW
  bnx2x: Fix no link on 577xx 10G-baseT
  bnx2x: Fix unrecognized SFP+ module after driver is loaded
  bnx2x: Fix potential incorrect link speed provision
  bnx2x: Restore global registers back to default.
  bnx2x: Fix link down in 57712 following LFA
  bnx2x: Fix 57810 1G-KR link against certain switches.
  ixgbe: PTP get_ts_info missing software support
  ...

12 years agotcp-repair: Handle zero-length data put in rcv queue
Pavel Emelyanov [Mon, 29 Oct 2012 05:05:33 +0000 (05:05 +0000)]
tcp-repair: Handle zero-length data put in rcv queue

When sending data into a tcp socket in repair state we should check
for the amount of data being 0 explicitly. Otherwise we'll have an skb
with seq == end_seq in rcv queue, but tcp doesn't expect this to happen
(in particular a warn_on in tcp_recvmsg shoots).

Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Reported-by: Giorgos Mavrikas <gmavrikas@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agovmxnet3: must split too big fragments
Eric Dumazet [Mon, 29 Oct 2012 07:30:49 +0000 (07:30 +0000)]
vmxnet3: must split too big fragments

vmxnet3 has a 16Kbytes limit per tx descriptor, that happened to work
as long as we provided PAGE_SIZE fragments.

Our stack can now build larger fragments, so we need to split them to
the 16kbytes boundary.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: jongman heo <jongman.heo@samsung.com>
Tested-by: jongman heo <jongman.heo@samsung.com>
Cc: Shreyas Bhatewara <sbhatewara@vmware.com>
Reviewed-by: Bhavesh Davda <bhavesh@vmware.com>
Signed-off-by: Shreyas Bhatewara <sbhatewara@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agol2tp: fix oops in l2tp_eth_create() error path
Tom Parkin [Mon, 29 Oct 2012 23:41:48 +0000 (23:41 +0000)]
l2tp: fix oops in l2tp_eth_create() error path

When creating an L2TPv3 Ethernet session, if register_netdev() should fail for
any reason (for example, automatic naming for "l2tpeth%d" interfaces hits the
32k-interface limit), the netdev is freed in the error path.  However, the
l2tp_eth_sess structure's dev pointer is left uncleared, and this results in
l2tp_eth_delete() then attempting to unregister the same netdev later in the
session teardown.  This results in an oops.

To avoid this, clear the session dev pointer in the error path.

Signed-off-by: Tom Parkin <tparkin@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agoexynos4_tmu_driver_ids should be exynos_tmu_driver_ids.
Jonghwan Choi [Tue, 23 Oct 2012 06:54:42 +0000 (14:54 +0800)]
exynos4_tmu_driver_ids should be exynos_tmu_driver_ids.

Signed-off-by: Jonghwan Choi <jhbird.choi@samsung.com>
Reviewed-by: Amit Daniel Kachhap <amit.kachhap@linaro.org>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
12 years agoACPI video: Ignore errors after _DOD evaluation.
Igor Murzov [Sat, 13 Oct 2012 00:41:25 +0000 (04:41 +0400)]
ACPI video: Ignore errors after _DOD evaluation.

There are systems where video module known to work fine regardless
of broken _DOD and ignoring returned value here doesn't cause
any issues later. This should fix brightness controls on some laptops.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=47861

Signed-off-by: Igor Murzov <e-mail@date.by>
Reviewed-by: Sergey V <sftp.mtuci@gmail.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
12 years agothermal: solve compilation errors in rcar_thermal
Devendra Naga [Wed, 31 Oct 2012 08:46:10 +0000 (17:46 +0900)]
thermal: solve compilation errors in rcar_thermal

following were the errors reported

drivers/thermal/rcar_thermal.c: In function â€˜rcar_thermal_probe’:
drivers/thermal/rcar_thermal.c:214:10: warning: passing argument 3 of â€˜thermal_zone_device_register’ makes integer from pointer without a cast [enabled by default]
include/linux/thermal.h:166:29: note: expected â€˜int’ but argument is of type â€˜struct rcar_thermal_priv *’
drivers/thermal/rcar_thermal.c:214:10: error: too few arguments to function â€˜thermal_zone_device_register’
include/linux/thermal.h:166:29: note: declared here
make[1]: *** [drivers/thermal/rcar_thermal.o] Error 1
make: *** [drivers/thermal/rcar_thermal.o] Error 2

with gcc version 4.6.3 (Ubuntu/Linaro 4.6.3-1ubuntu5)

Signed-off-by: Devendra Naga <develkernel412222@gmail.com>
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
12 years agocxgb4: Fix unable to get UP event from the LLD
Vipul Pandya [Mon, 29 Oct 2012 02:02:36 +0000 (02:02 +0000)]
cxgb4: Fix unable to get UP event from the LLD

If T4 configuration file gets loaded from the /lib/firmware/cxgb4/ directory
then offload capabilities of the cards were getting disabled during
initialization. Hence ULDs do not get an UP event from the LLD.

Signed-off-by: Jay Hernandez <jay@chelsio.com>
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agodrivers/net/phy/mdio-bitbang.c: Call mdiobus_unregister before mdiobus_free
Peter Senna Tschudin [Sun, 28 Oct 2012 06:12:01 +0000 (06:12 +0000)]
drivers/net/phy/mdio-bitbang.c: Call mdiobus_unregister before mdiobus_free

Based on commit b27393aecf66199f5ddad37c302d3e0cfadbe6c0

Calling mdiobus_free without calling mdiobus_unregister causes
BUG_ON(). This patch fixes the issue.

The semantic patch that found this issue(http://coccinelle.lip6.fr/):
// <smpl>
@@
expression E;
@@
  ... when != mdiobus_unregister(E);

+ mdiobus_unregister(E);
  mdiobus_free(E);
// </smpl>

Signed-off-by: Peter Senna Tschudin <peter.senna@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agodrivers/net/ethernet/nxp/lpc_eth.c: Call mdiobus_unregister before mdiobus_free
Peter Senna Tschudin [Sun, 28 Oct 2012 06:12:00 +0000 (06:12 +0000)]
drivers/net/ethernet/nxp/lpc_eth.c: Call mdiobus_unregister before mdiobus_free

Based on commit b27393aecf66199f5ddad37c302d3e0cfadbe6c0

Calling mdiobus_free without calling mdiobus_unregister causes
BUG_ON(). This patch fixes the issue.

The semantic patch that found this issue(http://coccinelle.lip6.fr/):
// <smpl>
@@
expression E;
@@
  ... when != mdiobus_unregister(E);

+ mdiobus_unregister(E);
  mdiobus_free(E);
// </smpl>

Signed-off-by: Peter Senna Tschudin <peter.senna@gmail.com>
Tested-by: Roland Stigge <stigge@antcom.de>
Tested-by: Alexandre Pereira da Silva <aletes.xgr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agobnx2x: fix HW initialization using fw 7.8.x
Dmitry Kravkov [Sun, 28 Oct 2012 21:59:04 +0000 (21:59 +0000)]
bnx2x: fix HW initialization using fw 7.8.x

Since commit 96bed4b9 (use FW 7.8.2) BRB HW block needs to be
initialized using fw values for all devices.
Otherwise ETS on 57712/578xx will not work.

Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agoMerge tag 'pm-for-3.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Fri, 2 Nov 2012 23:56:39 +0000 (16:56 -0700)]
Merge tag 'pm-for-3.7-rc4' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management update from Rafael J. Wysocki:
 "Change the email address of the powernow-k8 maintainer."

* tag 'pm-for-3.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq / powernow-k8: Change maintainer's email address

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Fri, 2 Nov 2012 23:11:15 +0000 (16:11 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input subsystem fixes from Dmitry Torokhov:
 "Just a few driver fixes."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: tsc40 - remove wrong announcement of pressure support
  Input: lpc32xx-keys - select INPUT_MATRIXKMAP
  Input: pxa27x_keypad - clear pending interrupts on keypad config
  Input: wacom - correct bad Cintiq 24HD check
  Input: wacom - add INPUT_PROP_DIRECT flag to Cintiq 24HD
  Input: egalax_ts - get gpio from devicetree

12 years agoNFS4: nfs4_opendata_access should return errno
Weston Andros Adamson [Fri, 2 Nov 2012 22:00:56 +0000 (18:00 -0400)]
NFS4: nfs4_opendata_access should return errno

Return errno - not an NFS4ERR_. This worked because NFS4ERR_ACCESS == EACCES.

Signed-off-by: Weston Andros Adamson <dros@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending
Linus Torvalds [Fri, 2 Nov 2012 22:08:20 +0000 (15:08 -0700)]
Merge git://git./linux/kernel/git/nab/target-pending

Pull more scsi target fixes from Nicholas Bellinger:
 "This series is a second round of target fixes for v3.7-rc4 that have
  come into target-devel over the last days, and are important enough to
  be applied ASAP.

  All are being CC'ed to stable.  The most important two are:

   - target: Re-add explict zeroing of INQUIRY bounce buffer memory to
     fix a regression for handling zero-length payloads, a bug that went
     during v3.7-rc1, and hit >= v3.6.3 stable.  (nab + paolo)

   - iscsi-target: Fix a long-standing missed R2T wakeup race in TX
     thread processing when using a single queue slot.  (Roland)

  Thanks to Roland & PureStorage team for helping to track down this
  long standing race with iscsi-target single queue slot operation.

  Also, the tcm_fc(FCoE) regression bug that was observed recently with
  -rc2 code has also been resolved with the cancel_delayed_work() return
  bugfix (commit c0158ca64da5: "workqueue: cancel_delayed_work() should
  return %false if work item is idle") now in -rc3.  Thanks again to Yi
  Zou, MDR, Robert Love @ Intel for helping to track this down."

* git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
  target: Fix incorrect usage of nested IRQ spinlocks in ABORT_TASK path
  iscsi-target: Fix missed wakeup race in TX thread
  target: Avoid integer overflow in se_dev_align_max_sectors()
  target: Don't return success from module_init() if setup fails
  target: Re-add explict zeroing of INQUIRY bounce buffer memory

12 years agoMerge tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck...
Linus Torvalds [Fri, 2 Nov 2012 20:27:52 +0000 (13:27 -0700)]
Merge tag 'hwmon-for-linus' of git://git./linux/kernel/git/groeck/linux-staging

Pull hwmon fixes from Guenter Roeck:
 "An e-mail address update, and fix a compile error on SPARC"

* tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: Only include of_match_table with CONFIG_OF_GPIO
  hwmon, fam15h_power: Change email address, MAINTAINERS entry

12 years agoMerge tag 'frv-fixes-20121102' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowe...
Linus Torvalds [Fri, 2 Nov 2012 20:27:01 +0000 (13:27 -0700)]
Merge tag 'frv-fixes-20121102' of git://git./linux/kernel/git/dhowells/linux-frv

Pull FRV fixes from David Howells:
 "A collection of small fixes for the FRV architecture."

* tag 'frv-fixes-20121102' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-frv:
  frv: fix the broken preempt
  frv: switch to saner kernel_execve() semantics
  FRV: Fix the new-style kernel_thread() stuff
  FRV: Fix the preemption handling
  FRV: gcc-4.1.2 also inlines weak functions
  FRV: Don't objcopy the GNU build_id note
  FRV: Add missing linux/export.h #inclusions

12 years agoMerge tag 'stable/for-linus-3.7-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 2 Nov 2012 20:26:11 +0000 (13:26 -0700)]
Merge tag 'stable/for-linus-3.7-rc4-tag' of git://git./linux/kernel/git/konrad/xen

Pull Xen bugfixes from Konrad Rzeszutek Wilk:
 - Use appropriate macros instead of hand-rolling our own (ARM).
 - Fixes if FB/KBD closed unexpectedly.
 - Fix memory leak in /dev/gntdev ioctl calls.
 - Fix overflow check in xenbus_file_write.
 - Document cleanup.
 - Performance optimization when migrating guests.

* tag 'stable/for-linus-3.7-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
  xen/mmu: Use Xen specific TLB flush instead of the generic one.
  xen/arm: use the __HVC macro
  xen/xenbus: fix overflow check in xenbus_file_write()
  xen-kbdfront: handle backend CLOSED without CLOSING
  xen-fbfront: handle backend CLOSED without CLOSING
  xen/gntdev: don't leak memory from IOCTL_GNTDEV_MAP_GRANT_REF
  x86: remove obsolete comment from asm/xen/hypervisor.h

12 years agohashtable: introduce a small and naive hashtable
Sasha Levin [Tue, 30 Oct 2012 18:45:57 +0000 (14:45 -0400)]
hashtable: introduce a small and naive hashtable

This hashtable implementation is using hlist buckets to provide a simple
hashtable to prevent it from getting reimplemented all over the kernel.

Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
[ Merging this now, so that subsystems can start applying Sasha's
  patches that use this   - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agofrv: fix the broken preempt
Al Viro [Fri, 2 Nov 2012 16:05:44 +0000 (12:05 -0400)]
frv: fix the broken preempt

Just get %icc2 into the state we would have after local_irq_disable()
and physical IRQ having happened since then.  Then we can simply
use preempt_schedule_irq() and be done with the whole mess.

Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
12 years agofrv: switch to saner kernel_execve() semantics
Al Viro [Mon, 15 Oct 2012 14:53:17 +0000 (10:53 -0400)]
frv: switch to saner kernel_execve() semantics

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
12 years agoFRV: Fix the new-style kernel_thread() stuff
David Howells [Fri, 2 Nov 2012 13:20:43 +0000 (13:20 +0000)]
FRV: Fix the new-style kernel_thread() stuff

The kernel_thread() changes for FRV don't work, and FRV fails to boot,
starting with:

commit 02ce496f152df87be081a64796498942c433a2fd
Author: Al Viro <viro@zeniv.linux.org.uk>
Date:   Tue Sep 18 22:18:51 2012 -0400
Subject: frv: split ret_from_fork, simplify kernel_thread() a lot

The problem is that the userspace registers are completely cleared when a
kernel thread is created and all subsequent user threads are then copied from
that.  Unfortunately, however, the TBR and PSR registers are restored from the
pt_regs and the values they should be set to are clobbered by the memset.

Instead, copy across the old user registers as normal, and then merely alter
GR8 and GR9 in it if we're going to execute a kernel thread.

Signed-off-by: David Howells <dhowells@redhat.com>
12 years agoFRV: Fix the preemption handling
David Howells [Fri, 2 Nov 2012 13:20:42 +0000 (13:20 +0000)]
FRV: Fix the preemption handling

Fix the preemption handling in FRV code where the PREEMPT_ACTIVE value is
incorrectly loaded into the threadinfo flags rather than the threadinfo
preemption count.

Unfortunately, the code cannot be simply converted to use
preempt_schedule_irq() as is because FRV uses virtual interrupt disablement to
cut down on the cost of actually disabling interrupts and thus
local_irq_enable() doesn't actually enable interrupts.

Reported-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Al Viro <viro@ZenIV.linux.org.uk>

12 years agoFRV: gcc-4.1.2 also inlines weak functions
David Howells [Fri, 2 Nov 2012 13:20:42 +0000 (13:20 +0000)]
FRV: gcc-4.1.2 also inlines weak functions

gcc-4.1.2 inlines weak functions, which causes FRV to fail when the dummy
thread_info_cache_init() gets inlined into start_kernel().

Signed-off-by: David Howells <dhowells@redhat.com>
12 years agoFRV: Don't objcopy the GNU build_id note
David Howells [Fri, 2 Nov 2012 13:20:42 +0000 (13:20 +0000)]
FRV: Don't objcopy the GNU build_id note

Don't let objcopy transfer the GNU build_id note into the loadable image as it
is located at address 0 and the image ends up >3G in size.

Signed-off-by: David Howells <dhowells@redhat.com>
12 years agoFRV: Add missing linux/export.h #inclusions
David Howells [Fri, 2 Nov 2012 13:20:42 +0000 (13:20 +0000)]
FRV: Add missing linux/export.h #inclusions

Add missing linux/export.h #inclusions to the FRV arch.

Signed-off-by: David Howells <dhowells@redhat.com>
12 years agoi2c: tegra: set irq name as device name
Laxman Dewangan [Thu, 1 Nov 2012 16:38:14 +0000 (22:08 +0530)]
i2c: tegra: set irq name as device name

When watching the irqs name of tegra i2c, all instances
irq name shows as tegra_i2c.

Passing the device name properly to have the irq names with
instance like tegra-i2c.0, tegra-i2c.1 etc.

Signed-off-by: Laxman Dewangan <ldewangan@nvidia.com>
Acked-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Wolfram Sang <w.sang@pengutronix.de>
12 years agoi2c-nomadik: Fixup clock handling
Philippe Begnic [Wed, 10 Oct 2012 11:02:26 +0000 (13:02 +0200)]
i2c-nomadik: Fixup clock handling

Make sure to clk_prepare as well as clk_enable.

Signed-off-by: Philippe Begnic <philippe.begnic@stericsson.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Wolfram Sang <w.sang@pengutronix.de>
12 years agoi2c: mxs: remove broken PIOQUEUE support
Wolfram Sang [Fri, 12 Oct 2012 10:55:16 +0000 (11:55 +0100)]
i2c: mxs: remove broken PIOQUEUE support

This I2C master can do DMA and PIOQUEUE (PIO with FIFO). Originally,
only PIOQUEUE was supported and it had issues, then DMA support was added
this cycle. The original intention was to keep PIOQUEUE since it has
less overhead what is nice for small transfers. However, runtime
switching between PIOQEUE and DMA depending on the transfer size never
worked despite a lot of trying. Since PIOQUEUE mode itself was flaky
(polling at places where interrupts failed to work) and the
implementation also imposed a size limit for transfers, it is best to
remove the support, so users don't fall over its limitations. It also
makes the driver a lot cleaner and more robust. If somebody really wants
less overhead, plain PIO mode could still be implemented with the
addidtional advantage that this mode is also available on MX23, too.

Signed-off-by: Wolfram Sang <w.sang@pengutronix.de>
Reviewed-by: Marek Vasut <marex@denx.de>
12 years agoMerge tag 'xtensa-next-20121101' of git://github.com/czankel/xtensa-linux
Linus Torvalds [Fri, 2 Nov 2012 00:48:19 +0000 (17:48 -0700)]
Merge tag 'xtensa-next-20121101' of git://github.com/czankel/xtensa-linux

Pull Xtensa fixes from Chris Zankel:
 "Some important bug fixes.

  With the change to uapi, there was a bug introduced that results in an
  empty syscall table (mult-inclusion bug).  Switching to the generic
  thread/execve allowed us to fix a bug we had in vfork()."

* tag 'xtensa-next-20121101' of git://github.com/czankel/xtensa-linux:
  xtensa: switch to generic sys_execve()
  xtensa: switch to generic kernel_execve()
  xtensa: switch to generic kernel_thread()
  xtensa: reset windowbase/windowstart when cloning the VM
  xtensa: use physical addresses for bus addresses
  xtensa: allow multi-inclusion for uapi/unistd.h

12 years agodrm/udl: fix stride issues scanning out stride != width*bpp
Dave Airlie [Thu, 1 Nov 2012 03:47:09 +0000 (13:47 +1000)]
drm/udl: fix stride issues scanning out stride != width*bpp

When buffer sharing with the i915 and using a 1680x1050 monitor,
the i915 gives is a 6912 buffer for the 6720 width, the code doesn't
render this properly as it uses one value to set the base address for
reading from the vmap and for where to start on the device.

This fixes it by calculating the values correctly for the device and
for the pixmap. No idea how I haven't seen this before now.

Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agohwmon: Only include of_match_table with CONFIG_OF_GPIO
Jamie Lentin [Thu, 1 Nov 2012 23:55:43 +0000 (23:55 +0000)]
hwmon: Only include of_match_table with CONFIG_OF_GPIO

The following fixes build errors on sparc. Without any DT support,
of_match_ptr is NULL and the below is a no-op. However, if just
CONFIG_OF is defined then so is of_match_ptr.

All useful parts of the gpio-fan DT support rely on CONFIG_OF_GPIO
anyway, so of_match_table should too.

Signed-off-by: Jamie Lentin <jm@lentin.co.uk>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
12 years agoMerge branch 'exynos-drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git...
Dave Airlie [Fri, 2 Nov 2012 00:30:37 +0000 (10:30 +1000)]
Merge branch 'exynos-drm-fixes' of git://git./linux/kernel/git/daeinki/drm-exynos into drm-fixes

Inki writes:
"As I posted before, we have added a new git repository for Exynos drm
to MAINTAINERS file so change it to new one like below,
   from git://git.infradead.org/users/kmpark/linux-samsung
   to git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos

And this pull request includes the following:
- fix display on issue when user requested dpms mode changing.
- add git repository for Exynos drm to MAINTAINERS file.
- add support for ARCH_MULTIPLATFORM.
- and code clean."

* 'exynos-drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
  drm: exynos: removed warning due to missing typecast for mixer driver data
  drm/exynos: add support for ARCH_MULTIPLATFORM
  MAINTAINERS: Add git repository for Exynos DRM
  drm/exynos: fix display on issue

12 years agoMerge branch 'drm-fixes-3.7' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Fri, 2 Nov 2012 00:29:22 +0000 (10:29 +1000)]
Merge branch 'drm-fixes-3.7' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

Alex writes:
"This request is mostly load detection fixes from Egbert and me."

* 'drm-fixes-3.7' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon: add load detection support for ext DAC on R200 (v2)
  DRM/radeon: For single CRTC GPUs move handling of CRTC_CRT_ON to crtc_dpms().
  DRM/Radeon: Fix TV DAC Load Detection for single CRTC chips.
  DRM/Radeon: Clean up code in TV DAC load detection.
  drm/radeon: fix ATPX function documentation
  drivers/gpu/drm/radeon/evergreen_cs.c: Remove unnecessary semicolon
  DRM/Radeon: On DVI-I use Load Detection when EDID is bogus.
  DRM/Radeon: Fix primary DAC Load Detection for RV100 chips.
  DRM/Radeon: Fix Load Detection on legacy primary DAC.

12 years agoMerge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel...
Dave Airlie [Fri, 2 Nov 2012 00:26:03 +0000 (10:26 +1000)]
Merge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel into drm-fixes

Daniel Vetter writes"
Nothing big at all for -fixes, just small stuff:
- Two patches to fix bugs on i830M
- ums regression fixer due to kicking firmeware fbs (Chris)
- tune down a too loud warning (Jani)
- be more careful with sdvo infoframes, which fixes a long-standing
  sdvo-hdmi regression"

* 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel:
  drm/i915: Only kick out vesafb if we takeover the fbcon with KMS
  drm/i915: be less verbose about inability to provide vendor backlight
  drm/i915: clear the entire sdvo infoframe buffer
  drm/i915: VGA needs to be on pipe A on i830M
  drm/i915: fix overlay on i830M

12 years agoNFSv4: Initialise the NFSv4.1 slot table highest_used_slotid correctly
Trond Myklebust [Mon, 29 Oct 2012 20:48:40 +0000 (16:48 -0400)]
NFSv4: Initialise the NFSv4.1 slot table highest_used_slotid correctly

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
12 years agotcp: Fix double sizeof in new tcp_metrics code
Julian Anastasov [Tue, 30 Oct 2012 12:03:09 +0000 (12:03 +0000)]
tcp: Fix double sizeof in new tcp_metrics code

Fix double sizeof when parsing IPv6 address from
user space because it breaks get/del by specific IPv6 address.

Problem noticed by David Binderman:

https://bugzilla.kernel.org/show_bug.cgi?id=49171

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agonet: fix divide by zero in tcp algorithm illinois
Jesper Dangaard Brouer [Wed, 31 Oct 2012 02:45:32 +0000 (02:45 +0000)]
net: fix divide by zero in tcp algorithm illinois

Reading TCP stats when using TCP Illinois congestion control algorithm
can cause a divide by zero kernel oops.

The division by zero occur in tcp_illinois_info() at:
 do_div(t, ca->cnt_rtt);
where ca->cnt_rtt can become zero (when rtt_reset is called)

Steps to Reproduce:
 1. Register tcp_illinois:
     # sysctl -w net.ipv4.tcp_congestion_control=illinois
 2. Monitor internal TCP information via command "ss -i"
     # watch -d ss -i
 3. Establish new TCP conn to machine

Either it fails at the initial conn, or else it needs to wait
for a loss or a reset.

This is only related to reading stats.  The function avg_delay() also
performs the same divide, but is guarded with a (ca->cnt_rtt > 0) at its
calling point in update_params().  Thus, simply fix tcp_illinois_info().

Function tcp_illinois_info() / get_info() is called without
socket lock.  Thus, eliminate any race condition on ca->cnt_rtt
by using a local stack variable.  Simply reuse info.tcpv_rttcnt,
as its already set to ca->cnt_rtt.
Function avg_delay() is not affected by this race condition, as
its called with the socket lock.

Cc: Petr Matousek <pmatouse@redhat.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agonet: sctp: Fix typo in net/sctp
Masanari Iida [Wed, 31 Oct 2012 05:48:19 +0000 (05:48 +0000)]
net: sctp: Fix typo in net/sctp

Correct spelling typo in net/sctp/socket.c

Signed-off-by: Masanari Iida <standby24x7@gmail.com>
Acked-by: Vlad Yasevich <vyasevich@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>