Currently to allocate a page that should be charged to kmemcg (e.g.
threadinfo), we pass __GFP_KMEMCG flag to the page allocator. The page
allocated is then to be freed by free_memcg_kmem_pages. Apart from
looking asymmetrical, this also requires intrusion to the general
allocation path. So let's introduce separate functions that will
alloc/free pages charged to kmemcg.
The new functions are called alloc_kmem_pages and free_kmem_pages. They
should be used when the caller actually would like to use kmalloc, but
has to fall back to the page allocator for the allocation is large.
They only differ from alloc_pages and free_pages in that besides
allocating or freeing pages they also charge them to the kmem resource
counter of the current memory cgroup.
[sfr@canb.auug.org.au: export kmalloc_order() to modules]
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Acked-by: Greg Thelen <gthelen@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Glauber Costa <glommer@gmail.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To increase compiler portability there is <linux/compiler.h> which
provides convenience macros for various gcc constructs. Eg: __weak for
__attribute__((weak)). I've replaced all instances of gcc attributes
with the right macro in the kernel subsystem.
Signed-off-by: Gideon Israel Dsouza <gidisrael@gmail.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
PF_MEMPOLICY is an unnecessary optimization for CONFIG_SLAB users.
There's no significant performance degradation to checking
current->mempolicy rather than current->flags & PF_MEMPOLICY in the
allocation path, especially since this is considered unlikely().
Running TCP_RR with netperf-2.4.5 through localhost on 16 cpu machine with
64GB of memory and without a mempolicy:
threads before after
16 1249409 1244487
32 1281786 1246783
48 1239175 1239138
64 1244642 1241841
80 1244346 1248918
96 1266436 1254316
112 1307398 1312135
128 1327607 1326502
Per-process flags are a scarce resource so we should free them up whenever
possible and make them available. We'll be using it shortly for memcg oom
reserves.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Jianguo Wu <wujianguo@huawei.com>
Cc: Tim Hockin <thockin@google.com>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch is a continuation of efforts trying to optimize find_vma(),
avoiding potentially expensive rbtree walks to locate a vma upon faults.
The original approach (https://lkml.org/lkml/2013/11/1/410), where the
largest vma was also cached, ended up being too specific and random,
thus further comparison with other approaches were needed. There are
two things to consider when dealing with this, the cache hit rate and
the latency of find_vma(). Improving the hit-rate does not necessarily
translate in finding the vma any faster, as the overhead of any fancy
caching schemes can be too high to consider.
We currently cache the last used vma for the whole address space, which
provides a nice optimization, reducing the total cycles in find_vma() by
up to 250%, for workloads with good locality. On the other hand, this
simple scheme is pretty much useless for workloads with poor locality.
Analyzing ebizzy runs shows that, no matter how many threads are
running, the mmap_cache hit rate is less than 2%, and in many situations
below 1%.
The proposed approach is to replace this scheme with a small per-thread
cache, maximizing hit rates at a very low maintenance cost.
Invalidations are performed by simply bumping up a 32-bit sequence
number. The only expensive operation is in the rare case of a seq
number overflow, where all caches that share the same address space are
flushed. Upon a miss, the proposed replacement policy is based on the
page number that contains the virtual address in question. Concretely,
the following results are seen on an 80 core, 8 socket x86-64 box:
1) System bootup: Most programs are single threaded, so the per-thread
scheme does improve ~50% hit rate by just adding a few more slots to
the cache.
+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline | 50.61% | 19.90 |
| patched | 73.45% | 13.58 |
+----------------+----------+------------------+
2) Kernel build: This one is already pretty good with the current
approach as we're dealing with good locality.
+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline | 75.28% | 11.03 |
| patched | 88.09% | 9.31 |
+----------------+----------+------------------+
3) Oracle 11g Data Mining (4k pages): Similar to the kernel build workload.
+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline | 70.66% | 17.14 |
| patched | 91.15% | 12.57 |
+----------------+----------+------------------+
4) Ebizzy: There's a fair amount of variation from run to run, but this
approach always shows nearly perfect hit rates, while baseline is just
about non-existent. The amounts of cycles can fluctuate between
anywhere from ~60 to ~116 for the baseline scheme, but this approach
reduces it considerably. For instance, with 80 threads:
+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline | 1.06% | 91.54 |
| patched | 99.97% | 14.18 |
+----------------+----------+------------------+
[akpm@linux-foundation.org: fix nommu build, per Davidlohr]
[akpm@linux-foundation.org: document vmacache_valid() logic]
[akpm@linux-foundation.org: attempt to untangle header files]
[akpm@linux-foundation.org: add vmacache_find() BUG_ON]
[hughd@google.com: add vmacache_valid_mm() (from Oleg)]
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: adjust and enhance comments]
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Michel Lespinasse <walken@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Tested-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull cgroup updates from Tejun Heo:
"A lot updates for cgroup:
- The biggest one is cgroup's conversion to kernfs. cgroup took
after the long abandoned vfs-entangled sysfs implementation and
made it even more convoluted over time. cgroup's internal objects
were fused with vfs objects which also brought in vfs locking and
object lifetime rules. Naturally, there are places where vfs rules
don't fit and nasty hacks, such as credential switching or lock
dance interleaving inode mutex and cgroup_mutex with object serial
number comparison thrown in to decide whether the operation is
actually necessary, needed to be employed.
After conversion to kernfs, internal object lifetime and locking
rules are mostly isolated from vfs interactions allowing shedding
of several nasty hacks and overall simplification. This will also
allow implmentation of operations which may affect multiple cgroups
which weren't possible before as it would have required nesting
i_mutexes.
- Various simplifications including dropping of module support,
easier cgroup name/path handling, simplified cgroup file type
handling and task_cg_lists optimization.
- Prepatory changes for the planned unified hierarchy, which is still
a patchset away from being actually operational. The dummy
hierarchy is updated to serve as the default unified hierarchy.
Controllers which aren't claimed by other hierarchies are
associated with it, which BTW was what the dummy hierarchy was for
anyway.
- Various fixes from Li and others. This pull request includes some
patches to add missing slab.h to various subsystems. This was
triggered xattr.h include removal from cgroup.h. cgroup.h
indirectly got included a lot of files which brought in xattr.h
which brought in slab.h.
There are several merge commits - one to pull in kernfs updates
necessary for converting cgroup (already in upstream through
driver-core), others for interfering changes in the fixes branch"
* 'for-3.15' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup: (74 commits)
cgroup: remove useless argument from cgroup_exit()
cgroup: fix spurious lockdep warning in cgroup_exit()
cgroup: Use RCU_INIT_POINTER(x, NULL) in cgroup.c
cgroup: break kernfs active_ref protection in cgroup directory operations
cgroup: fix cgroup_taskset walking order
cgroup: implement CFTYPE_ONLY_ON_DFL
cgroup: make cgrp_dfl_root mountable
cgroup: drop const from @buffer of cftype->write_string()
cgroup: rename cgroup_dummy_root and related names
cgroup: move ->subsys_mask from cgroupfs_root to cgroup
cgroup: treat cgroup_dummy_root as an equivalent hierarchy during rebinding
cgroup: remove NULL checks from [pr_cont_]cgroup_{name|path}()
cgroup: use cgroup_setup_root() to initialize cgroup_dummy_root
cgroup: reorganize cgroup bootstrapping
cgroup: relocate setting of CGRP_DEAD
cpuset: use rcu_read_lock() to protect task_cs()
cgroup_freezer: document freezer_fork() subtleties
cgroup: update cgroup_transfer_tasks() to either succeed or fail
cgroup: drop task_lock() protection around task->cgroups
cgroup: update how a newly forked task gets associated with css_set
...
cgroup_exit() is called in fork and exit path. If it's called in the
failure path during fork, PF_EXITING isn't set, and then lockdep will
complain.
Fix this by removing cgroup_exit() in that failure path. cgroup_fork()
does nothing that needs cleanup.
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
current->mm doesn't need a NULL check in dup_mm(). Becasue dup_mm() is
used only in copy_mm() and current->mm is checked whether it is NULL or
not in copy_mm() before calling dup_mm().
Signed-off-by: Daeseok Youn <daeseok.youn@gmail.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
while_each_thread() and next_thread() should die, almost every lockless
usage is wrong.
1. Unless g == current, the lockless while_each_thread() is not safe.
while_each_thread(g, t) can loop forever if g exits, next_thread()
can't reach the unhashed thread in this case. Note that this can
happen even if g is the group leader, it can exec.
2. Even if while_each_thread() itself was correct, people often use
it wrongly.
It was never safe to just take rcu_read_lock() and loop unless
you verify that pid_alive(g) == T, even the first next_thread()
can point to the already freed/reused memory.
This patch adds signal_struct->thread_head and task->thread_node to
create the normal rcu-safe list with the stable head. The new
for_each_thread(g, t) helper is always safe under rcu_read_lock() as
long as this task_struct can't go away.
Note: of course it is ugly to have both task_struct->thread_node and the
old task_struct->thread_group, we will kill it later, after we change
the users of while_each_thread() to use for_each_thread().
Perhaps we can kill it even before we convert all users, we can
reimplement next_thread(t) using the new thread_head/thread_node. But
we can't do this right now because this will lead to subtle behavioural
changes. For example, do/while_each_thread() always sees at least one
task, while for_each_thread() can do nothing if the whole thread group
has died. Or thread_group_empty(), currently its semantics is not clear
unless thread_group_leader(p) and we need to audit the callers before we
can change it.
So this patch adds the new interface which has to coexist with the old
one for some time, hopefully the next changes will be more or less
straightforward and the old one will go away soon.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Sergey Dyasly <dserrg@gmail.com>
Tested-by: Sergey Dyasly <dserrg@gmail.com>
Reviewed-by: Sameer Nanda <snanda@chromium.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mandeep Singh Baines <msb@chromium.org>
Cc: "Ma, Xindong" <xindong.ma@intel.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: "Tu, Xiaobing" <xiaobing.tu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull scheduler changes from Ingo Molnar:
- Add the initial implementation of SCHED_DEADLINE support: a real-time
scheduling policy where tasks that meet their deadlines and
periodically execute their instances in less than their runtime quota
see real-time scheduling and won't miss any of their deadlines.
Tasks that go over their quota get delayed (Available to privileged
users for now)
- Clean up and fix preempt_enable_no_resched() abuse all around the
tree
- Do sched_clock() performance optimizations on x86 and elsewhere
- Fix and improve auto-NUMA balancing
- Fix and clean up the idle loop
- Apply various cleanups and fixes
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (60 commits)
sched: Fix __sched_setscheduler() nice test
sched: Move SCHED_RESET_ON_FORK into attr::sched_flags
sched: Fix up attr::sched_priority warning
sched: Fix up scheduler syscall LTP fails
sched: Preserve the nice level over sched_setscheduler() and sched_setparam() calls
sched/core: Fix htmldocs warnings
sched/deadline: No need to check p if dl_se is valid
sched/deadline: Remove unused variables
sched/deadline: Fix sparse static warnings
m68k: Fix build warning in mac_via.h
sched, thermal: Clean up preempt_enable_no_resched() abuse
sched, net: Fixup busy_loop_us_clock()
sched, net: Clean up preempt_enable_no_resched() abuse
sched/preempt: Fix up missed PREEMPT_NEED_RESCHED folding
sched/preempt, locking: Rework local_bh_{dis,en}able()
sched/clock, x86: Avoid a runtime condition in native_sched_clock()
sched/clock: Fix up clear_sched_clock_stable()
sched/clock, x86: Use a static_key for sched_clock_stable
sched/clock: Remove local_irq_disable() from the clocks
sched/clock, x86: Rewrite cyc2ns() to avoid the need to disable IRQs
...
Pull namespace fixes from Eric Biederman:
"This is a set of 3 regression fixes.
This fixes /proc/mounts when using "ip netns add <netns>" to display
the actual mount point.
This fixes a regression in clone that broke lxc-attach.
This fixes a regression in the permission checks for mounting /proc
that made proc unmountable if binfmt_misc was in use. Oops.
My apologies for sending this pull request so late. Al Viro gave
interesting review comments about the d_path fix that I wanted to
address in detail before I sent this pull request. Unfortunately a
bad round of colds kept from addressing that in detail until today.
The executive summary of the review was:
Al: Is patching d_path really sufficient?
The prepend_path, d_path, d_absolute_path, and __d_path family of
functions is a really mess.
Me: Yes, patching d_path is really sufficient. Yes, the code is mess.
No it is not appropriate to rewrite all of d_path for a regression
that has existed for entirely too long already, when a two line
change will do"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
vfs: Fix a regression in mounting proc
fork: Allow CLONE_PARENT after setns(CLONE_NEWPID)
vfs: In d_path don't call d_dname on a mount point
Some method to deal with rt-mutexes and make sched_dl interact with
the current PI-coded is needed, raising all but trivial issues, that
needs (according to us) to be solved with some restructuring of
the pi-code (i.e., going toward a proxy execution-ish implementation).
This is under development, in the meanwhile, as a temporary solution,
what this commits does is:
- ensure a pi-lock owner with waiters is never throttled down. Instead,
when it runs out of runtime, it immediately gets replenished and it's
deadline is postponed;
- the scheduling parameters (relative deadline and default runtime)
used for that replenishments --during the whole period it holds the
pi-lock-- are the ones of the waiting task with earliest deadline.
Acting this way, we provide some kind of boosting to the lock-owner,
still by using the existing (actually, slightly modified by the previous
commit) pi-architecture.
We would stress the fact that this is only a surely needed, all but
clean solution to the problem. In the end it's only a way to re-start
discussion within the community. So, as always, comments, ideas, rants,
etc.. are welcome! :-)
Signed-off-by: Dario Faggioli <raistlin@linux.it>
Signed-off-by: Juri Lelli <juri.lelli@gmail.com>
[ Added !RT_MUTEXES build fix. ]
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1383831828-15501-11-git-send-email-juri.lelli@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Turn the pi-chains from plist to rb-tree, in the rt_mutex code,
and provide a proper comparison function for -deadline and
-priority tasks.
This is done mainly because:
- classical prio field of the plist is just an int, which might
not be enough for representing a deadline;
- manipulating such a list would become O(nr_deadline_tasks),
which might be to much, as the number of -deadline task increases.
Therefore, an rb-tree is used, and tasks are queued in it according
to the following logic:
- among two -priority (i.e., SCHED_BATCH/OTHER/RR/FIFO) tasks, the
one with the higher (lower, actually!) prio wins;
- among a -priority and a -deadline task, the latter always wins;
- among two -deadline tasks, the one with the earliest deadline
wins.
Queueing and dequeueing functions are changed accordingly, for both
the list of a task's pi-waiters and the list of tasks blocked on
a pi-lock.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Dario Faggioli <raistlin@linux.it>
Signed-off-by: Juri Lelli <juri.lelli@gmail.com>
Signed-off-again-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1383831828-15501-10-git-send-email-juri.lelli@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Introduces the data structures, constants and symbols needed for
SCHED_DEADLINE implementation.
Core data structure of SCHED_DEADLINE are defined, along with their
initializers. Hooks for checking if a task belong to the new policy
are also added where they are needed.
Adds a scheduling class, in sched/dl.c and a new policy called
SCHED_DEADLINE. It is an implementation of the Earliest Deadline
First (EDF) scheduling algorithm, augmented with a mechanism (called
Constant Bandwidth Server, CBS) that makes it possible to isolate
the behaviour of tasks between each other.
The typical -deadline task will be made up of a computation phase
(instance) which is activated on a periodic or sporadic fashion. The
expected (maximum) duration of such computation is called the task's
runtime; the time interval by which each instance need to be completed
is called the task's relative deadline. The task's absolute deadline
is dynamically calculated as the time instant a task (better, an
instance) activates plus the relative deadline.
The EDF algorithms selects the task with the smallest absolute
deadline as the one to be executed first, while the CBS ensures each
task to run for at most its runtime every (relative) deadline
length time interval, avoiding any interference between different
tasks (bandwidth isolation).
Thanks to this feature, also tasks that do not strictly comply with
the computational model sketched above can effectively use the new
policy.
To summarize, this patch:
- introduces the data structures, constants and symbols needed;
- implements the core logic of the scheduling algorithm in the new
scheduling class file;
- provides all the glue code between the new scheduling class and
the core scheduler and refines the interactions between sched/dl
and the other existing scheduling classes.
Signed-off-by: Dario Faggioli <raistlin@linux.it>
Signed-off-by: Michael Trimarchi <michael@amarulasolutions.com>
Signed-off-by: Fabio Checconi <fchecconi@gmail.com>
Signed-off-by: Juri Lelli <juri.lelli@gmail.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1383831828-15501-4-git-send-email-juri.lelli@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There are a few subtle races, between change_protection_range (used by
mprotect and change_prot_numa) on one side, and NUMA page migration and
compaction on the other side.
The basic race is that there is a time window between when the PTE gets
made non-present (PROT_NONE or NUMA), and the TLB is flushed.
During that time, a CPU may continue writing to the page.
This is fine most of the time, however compaction or the NUMA migration
code may come in, and migrate the page away.
When that happens, the CPU may continue writing, through the cached
translation, to what is no longer the current memory location of the
process.
This only affects x86, which has a somewhat optimistic pte_accessible.
All other architectures appear to be safe, and will either always flush,
or flush whenever there is a valid mapping, even with no permissions
(SPARC).
The basic race looks like this:
CPU A CPU B CPU C
load TLB entry
make entry PTE/PMD_NUMA
fault on entry
read/write old page
start migrating page
change PTE/PMD to new page
read/write old page [*]
flush TLB
reload TLB from new entry
read/write new page
lose data
[*] the old page may belong to a new user at this point!
The obvious fix is to flush remote TLB entries, by making sure that
pte_accessible aware of the fact that PROT_NONE and PROT_NUMA memory may
still be accessible if there is a TLB flush pending for the mm.
This should fix both NUMA migration and compaction.
[mgorman@suse.de: fix build]
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Serge Hallyn <serge.hallyn@ubuntu.com> writes:
> Hi Oleg,
>
> commit 40a0d32d1e :
> "fork: unify and tighten up CLONE_NEWUSER/CLONE_NEWPID checks"
> breaks lxc-attach in 3.12. That code forks a child which does
> setns() and then does a clone(CLONE_PARENT). That way the
> grandchild can be in the right namespaces (which the child was
> not) and be a child of the original task, which is the monitor.
>
> lxc-attach in 3.11 was working fine with no side effects that I
> could see. Is there a real danger in allowing CLONE_PARENT
> when current->nsproxy->pidns_for_children is not our pidns,
> or was this done out of an "over-abundance of caution"? Can we
> safely revert that new extra check?
The two fundamental things I know we can not allow are:
- A shared signal queue aka CLONE_THREAD. Because we compute the pid
and uid of the signal when we place it in the queue.
- Changing the pid and by extention pid_namespace of an existing
process.
From a parents perspective there is nothing special about the pid
namespace, to deny CLONE_PARENT, because the parent simply won't know or
care.
From the childs perspective all that is special really are shared signal
queues.
User mode threading with CLONE_PARENT|CLONE_VM|CLONE_SIGHAND and tasks
in different pid namespaces is almost certainly going to break because
it is complicated. But shared signal handlers can look at per thread
information to know which pid namespace a process is in, so I don't know
of any reason not to support CLONE_PARENT|CLONE_VM|CLONE_SIGHAND threads
at the kernel level. It would be absolutely stupid to implement but
that is a different thing.
So hmm.
Because it can do no harm, and because it is a regression let's remove
the CLONE_PARENT check and send it stable.
Cc: stable@vger.kernel.org
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
