In order to remove dropping rq->lock from the
switched_{to,from}()/prio_changed() sched_class methods, run the
balance callbacks after it.
We need to remove dropping rq->lock because its buggy,
suppose using sched_setattr()/sched_setscheduler() to change a running
task from FIFO to OTHER.
By the time we get to switched_from_rt() the task is already enqueued
on the cfs runqueues. If switched_from_rt() does pull_rt_task() and
drops rq->lock, load-balancing can come in and move our task @p to
another rq.
The subsequent switched_to_fair() still assumes @p is on @rq and bad
things will happen.
By using balance callbacks we delay the load-balancing operations
{rt,dl}x{push,pull} until we've done all the important work and the
task is fully set up.
Furthermore, the balance callbacks do not know about @p, therefore
they cannot get confused like this.
Reported-by: Mike Galbraith <umgwanakikbuti@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: ktkhai@parallels.com
Cc: rostedt@goodmis.org
Cc: juri.lelli@gmail.com
Cc: pang.xunlei@linaro.org
Cc: oleg@redhat.com
Cc: wanpeng.li@linux.intel.com
Link: http://lkml.kernel.org/r/20150611124742.615343911@infradead.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
On traditional hierarchies, if a task has write access to "tasks" or
"cgroup.procs" file of a cgroup and its euid agrees with the target,
it can move the target to the cgroup; however, consider the following
scenario. The owner of each cgroup is in the parentheses.
R (root) - 0 (root) - 00 (user1) - 000 (user1)
| \ 001 (user1)
\ 1 (root) - 10 (user1)
The subtrees of 00 and 10 are delegated to user1; however, while both
subtrees may belong to the same user, it is clear that the two
subtrees are to be isolated - they're under completely separate
resource limits imposed by 0 and 1, respectively. Note that 0 and 1
aren't strictly necessary but added to ease illustrating the issue.
If user1 is allowed to move processes between the two subtrees, the
intention of the hierarchy - keeping a given group of processes under
a subtree with certain resource restrictions while delegating
management of the subtree - can be circumvented by user1.
This happens because migration permission check doesn't consider the
hierarchical nature of cgroups. To fix the issue, this patch adds an
extra permission requirement when userland tries to migrate a process
in the default hierarchy - the issuing task must have write access to
the common ancestor of "cgroup.procs" file of the ancestor in addition
to the destination's.
Conceptually, the issuer must be able to move the target process from
the source cgroup to the common ancestor of source and destination
cgroups and then to the destination. As long as delegation is done in
a proper top-down way, this guarantees that a delegatee can't smuggle
processes across disjoint delegation domains.
The next patch will add documentation on the delegation model on the
default hierarchy.
v2: Fixed missing !ret test. Spotted by Li Zefan.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Separate out task / process migration permission check from
__cgroup_procs_write() into cgroup_procs_write_permission().
* Permission check is moved right above the actual migration and no
longer performed while holding rcu_read_lock().
cgroup_procs_write_permission() uses get_task_cred() / put_cred()
instead of __task_cred(). Also, !root trying to migrate kthreadd or
PF_NO_SETAFFINITY tasks will now fail with -EINVAL rather than
-EACCES which should be fine.
* The same permission check is now performed even when moving self by
specifying 0 as pid. This always succeeds so there's no functional
difference. We'll add more permission checks later and the benefits
of keeping both cases consistent outweigh the minute overhead of
doing perm checks on pid 0 case.
Signed-off-by: Tejun Heo <tj@kernel.org>
Mark the task for later wakeup after the wait_lock has been released.
This way, once the next task is awoken, it will have a better chance
to of finding the wait_lock free when continuing executing in
__rt_mutex_slowlock() when trying to acquire the rtmutex, calling
try_to_take_rt_mutex(). Upon contended scenarios, other tasks attempting
take the lock may acquire it first, right after the wait_lock is released,
but (a) this can also occur with the current code, as it relies on the
spinlock fairness, and (b) we are dealing with the top-waiter anyway,
so it will always take the lock next.
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1432056298-18738-2-git-send-email-dave@stgolabs.net
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The fix in d151832650 (time: Move clock_was_set_seq update
before updating shadow-timekeeper) was unfortunately incomplete.
The main gist of that change was to do the shadow-copy update
last, so that any state changes were properly duplicated, and
we wouldn't accidentally have stale data in the shadow.
Unfortunately in the main update_wall_time() logic, we update
use the shadow-timekeeper to calculate the next update values,
then while holding the lock, copy the shadow-timekeeper over,
then call timekeeping_update() to do some additional
bookkeeping, (skipping the shadow mirror). The bug with this is
the additional bookkeeping isn't all read-only, and some
changes timkeeper state. Thus we might then overwrite this state
change on the next update.
To avoid this problem, do the timekeeping_update() on the
shadow-timekeeper prior to copying the full state over to
the real-timekeeper.
This avoids problems with both the clock_was_set_seq and
next_leap_ktime being overwritten and possibly the
fast-timekeepers as well.
Many thanks to Prarit for his rigorous testing, which discovered
this problem, along with Prarit and Daniel's work validating this
fix.
Reported-by: Prarit Bhargava <prarit@redhat.com>
Tested-by: Prarit Bhargava <prarit@redhat.com>
Tested-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jiri Bohac <jbohac@suse.cz>
Cc: Ingo Molnar <mingo@kernel.org>
Link: http://lkml.kernel.org/r/1434560753-7441-1-git-send-email-john.stultz@linaro.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Pull tracing filter fix from Steven Rostedt:
"Vince Weaver reported a warning when he added perf event filters into
his fuzzer tests. There's a missing check of balanced operations when
parenthesis are used, and this triggers a WARN_ON() and when reading
the failure, the filter reports no failure occurred.
The operands were not being checked if they match, this adds that"
* tag 'trace-fix-filter-4.1-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
tracing: Have filter check for balanced ops
When the following filter is used it causes a warning to trigger:
# cd /sys/kernel/debug/tracing
# echo "((dev==1)blocks==2)" > events/ext4/ext4_truncate_exit/filter
-bash: echo: write error: Invalid argument
# cat events/ext4/ext4_truncate_exit/filter
((dev==1)blocks==2)
^
parse_error: No error
------------[ cut here ]------------
WARNING: CPU: 2 PID: 1223 at kernel/trace/trace_events_filter.c:1640 replace_preds+0x3c5/0x990()
Modules linked in: bnep lockd grace bluetooth ...
CPU: 3 PID: 1223 Comm: bash Tainted: G W 4.1.0-rc3-test+ #450
Hardware name: Hewlett-Packard HP Compaq Pro 6300 SFF/339A, BIOS K01 v02.05 05/07/2012
0000000000000668 ffff8800c106bc98 ffffffff816ed4f9 ffff88011ead0cf0
0000000000000000 ffff8800c106bcd8 ffffffff8107fb07 ffffffff8136b46c
ffff8800c7d81d48 ffff8800d4c2bc00 ffff8800d4d4f920 00000000ffffffea
Call Trace:
[<ffffffff816ed4f9>] dump_stack+0x4c/0x6e
[<ffffffff8107fb07>] warn_slowpath_common+0x97/0xe0
[<ffffffff8136b46c>] ? _kstrtoull+0x2c/0x80
[<ffffffff8107fb6a>] warn_slowpath_null+0x1a/0x20
[<ffffffff81159065>] replace_preds+0x3c5/0x990
[<ffffffff811596b2>] create_filter+0x82/0xb0
[<ffffffff81159944>] apply_event_filter+0xd4/0x180
[<ffffffff81152bbf>] event_filter_write+0x8f/0x120
[<ffffffff811db2a8>] __vfs_write+0x28/0xe0
[<ffffffff811dda43>] ? __sb_start_write+0x53/0xf0
[<ffffffff812e51e0>] ? security_file_permission+0x30/0xc0
[<ffffffff811dc408>] vfs_write+0xb8/0x1b0
[<ffffffff811dc72f>] SyS_write+0x4f/0xb0
[<ffffffff816f5217>] system_call_fastpath+0x12/0x6a
---[ end trace e11028bd95818dcd ]---
Worse yet, reading the error message (the filter again) it says that
there was no error, when there clearly was. The issue is that the
code that checks the input does not check for balanced ops. That is,
having an op between a closed parenthesis and the next token.
This would only cause a warning, and fail out before doing any real
harm, but it should still not caues a warning, and the error reported
should work:
# cd /sys/kernel/debug/tracing
# echo "((dev==1)blocks==2)" > events/ext4/ext4_truncate_exit/filter
-bash: echo: write error: Invalid argument
# cat events/ext4/ext4_truncate_exit/filter
((dev==1)blocks==2)
^
parse_error: Meaningless filter expression
And give no kernel warning.
Link: http://lkml.kernel.org/r/20150615175025.7e809215@gandalf.local.home
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: stable@vger.kernel.org # 2.6.31+
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Tested-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
This lets you build a kernel which can support xen dom0
or xen guests on i386, x86-64 and arm64 by just using:
make xenconfig
You can start from an allnoconfig and then switch to xenconfig.
This also splits out the options which are available currently
to be built with x86 and 'make ARCH=arm64' under a shared config.
Technically xen supports a dom0 kernel and also a guest
kernel configuration but upon review with the xen team
since we don't have many dom0 options its best to just
combine these two into one.
A few generic notes: we enable both of these:
CONFIG_INET=y
CONFIG_BINFMT_ELF=y
although technically not required given you likely will
end up with a pretty useless system otherwise.
A few architectural differences worth noting:
$ make allnoconfig; make xenconfig > /dev/null ; \
grep XEN .config > 64-bit-config
$ make ARCH=i386 allnoconfig; make ARCH=i386 xenconfig > /dev/null; \
grep XEN .config > 32-bit-config
$ make ARCH=arm64 allnoconfig; make ARCH=arm64 xenconfig > /dev/null; \
grep XEN .config > arm64-config
Since the options are already split up with a generic config and
architecture specific configs you anything on the x86 configs
are known to only work right now on x86. For instance arm64 doesn't
support MEMORY_HOTPLUG yet as such although we try to enabe it
generically arm64 doesn't have it yet, so we leave the xen
specific kconfig option XEN_BALLOON_MEMORY_HOTPLUG on x86's config
file to set expecations correctly.
Then on x86 we have differences between i386 and x86-64. The difference
between 64-bit-config and 32-bit-config is you don't get XEN_MCE_LOG as
this is only supported on 64-bit. You also do not get on i386
XEN_BALLOON_MEMORY_HOTPLUG, there does not seem to be any technical
reasons to not allow this but I gave up after a few attempts.
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Borislav Petkov <bp@suse.de>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: penberg@kernel.org
Cc: levinsasha928@gmail.com
Cc: mtosatti@redhat.com
Cc: fengguang.wu@intel.com
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Ian Campbell <Ian.Campbell@citrix.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: xen-devel@lists.xenproject.org
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Acked-by: Julien Grall <julien.grall@linaro.org>
Acked-by: Michal Marek <mmarek@suse.cz>
Acked-by: David Rientjes <rientjes@google.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
bpf_trace_printk() is a helper function used to debug eBPF programs.
Let socket and TC programs use it as well.
Note, it's DEBUG ONLY helper. If it's used in the program,
the kernel will print warning banner to make sure users don't use
it in production.
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
eBPF programs attached to kprobes need to filter based on
current->pid, uid and other fields, so introduce helper functions:
u64 bpf_get_current_pid_tgid(void)
Return: current->tgid << 32 | current->pid
u64 bpf_get_current_uid_gid(void)
Return: current_gid << 32 | current_uid
bpf_get_current_comm(char *buf, int size_of_buf)
stores current->comm into buf
They can be used from the programs attached to TC as well to classify packets
based on current task fields.
Update tracex2 example to print histogram of write syscalls for each process
instead of aggregated for all.
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It takes a while until the ring_buffer_benchmark module is removed
when the ring buffer hammer is running. It is because it takes
few seconds and kthread_should_stop() is not being checked.
This patch adds the check for kthread termination into the producer.
It uses the existing @kill_test flag to finish the kthreads as
cleanly as possible.
It disables printing the "ERROR" message when the kthread is going.
It makes sure that producer does not go into the 10sec sleep
when it is being killed.
Finally, it does not call wait_to_die() when kthread_should_stop()
already returns true.
Link: http://lkml.kernel.org/r/20150615155428.GD3135@pathway.suse.cz
Signed-off-by: Petr Mladek <pmladek@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Pull lockdep fix from Ingo Molnar:
"A lockdep/modules unload race fix that can oops"
* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
lockdep: Fix a race between /proc/lock_stat and module unload
Since the leapsecond is applied at tick-time, this means there is a
small window of time at the start of a leap-second where we cross into
the next second before applying the leap.
This patch modified adjtimex so that the leap-second is applied on the
second edge. Providing more correct leapsecond behavior.
This does make it so that adjtimex()'s returned time values can be
inconsistent with time values read from gettimeofday() or
clock_gettime(CLOCK_REALTIME,...) for a brief period of one tick at
the leapsecond. However, those other interfaces do not provide the
TIME_OOP time_state return that adjtimex() provides, which allows the
leapsecond to be properly represented. They instead only see a time
discontinuity, and cannot tell the first 23:59:59 from the repeated
23:59:59 leap second.
This seems like a reasonable tradeoff given clock_gettime() /
gettimeofday() cannot properly represent a leapsecond, and users
likely care more about performance, while folks who are using
adjtimex() more likely care about leap-second correctness.
Signed-off-by: John Stultz <john.stultz@linaro.org>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jiri Bohac <jbohac@suse.cz>
Cc: Ingo Molnar <mingo@kernel.org>
Link: http://lkml.kernel.org/r/1434063297-28657-5-git-send-email-john.stultz@linaro.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Currently, leapsecond adjustments are done at tick time. As a result,
the leapsecond was applied at the first timer tick *after* the
leapsecond (~1-10ms late depending on HZ), rather then exactly on the
second edge.
This was in part historical from back when we were always tick based,
but correcting this since has been avoided since it adds extra
conditional checks in the gettime fastpath, which has performance
overhead.
However, it was recently pointed out that ABS_TIME CLOCK_REALTIME
timers set for right after the leapsecond could fire a second early,
since some timers may be expired before we trigger the timekeeping
timer, which then applies the leapsecond.
This isn't quite as bad as it sounds, since behaviorally it is similar
to what is possible w/ ntpd made leapsecond adjustments done w/o using
the kernel discipline. Where due to latencies, timers may fire just
prior to the settimeofday call. (Also, one should note that all
applications using CLOCK_REALTIME timers should always be careful,
since they are prone to quirks from settimeofday() disturbances.)
However, the purpose of having the kernel do the leap adjustment is to
avoid such latencies, so I think this is worth fixing.
So in order to properly keep those timers from firing a second early,
this patch modifies the ntp and timekeeping logic so that we keep
enough state so that the update_base_offsets_now accessor, which
provides the hrtimer core the current time, can check and apply the
leapsecond adjustment on the second edge. This prevents the hrtimer
core from expiring timers too early.
This patch does not modify any other time read path, so no additional
overhead is incurred. However, this also means that the leap-second
continues to be applied at tick time for all other read-paths.
Apologies to Richard Cochran, who pushed for similar changes years
ago, which I resisted due to the concerns about the performance
overhead.
While I suspect this isn't extremely critical, folks who care about
strict leap-second correctness will likely want to watch
this. Potentially a -stable candidate eventually.
Originally-suggested-by: Richard Cochran <richardcochran@gmail.com>
Reported-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Reported-by: Prarit Bhargava <prarit@redhat.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jiri Bohac <jbohac@suse.cz>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Cc: Ingo Molnar <mingo@kernel.org>
Link: http://lkml.kernel.org/r/1434063297-28657-4-git-send-email-john.stultz@linaro.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
It was reported that 868a3e915f (hrtimer: Make offset
update smarter) was causing timer problems after suspend/resume.
The problem with that change is the modification to
clock_was_set_seq in timekeeping_update is done prior to
mirroring the time state to the shadow-timekeeper. Thus the
next time we do update_wall_time() the updated sequence is
overwritten by whats in the shadow copy.
This patch moves the shadow-timekeeper mirroring to the end
of the function, after all updates have been made, so all data
is kept in sync.
(This patch also affects the update_fast_timekeeper calls which
were also problematically done prior to the mirroring).
Reported-and-tested-by: Jeremiah Mahler <jmmahler@gmail.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Cc: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/r/1434063297-28657-2-git-send-email-john.stultz@linaro.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Pull ring buffer benchmark buglet fix from Steven Rostedt:
"Wang Long fixed a minor bug in the module parameter for the ring
buffer benchmark, where the produce_fifo was being ignored and the
producer thread's priority was being set with the consumer_fifo
parameter"
* tag 'trace-rb-bm-fix-4.1-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
ring-buffer-benchmark: Fix the wrong sched_priority of producer
strnlen_user() returns 0 when it hits fault, not -1. Fix the test in
audit_log_single_execve_arg(). Luckily this shouldn't ever happen unless
there's a kernel bug so it's mostly a cosmetic fix.
CC: Paul Moore <pmoore@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Paul Moore <pmoore@redhat.com>
Jovi Zhangwei reported the following problem
Below kernel vm bug can be triggered by tcpdump which mmaped a lot of pages
with GFP_COMP flag.
[Mon May 25 05:29:33 2015] page:ffffea0015414000 count:66 mapcount:1 mapping: (null) index:0x0
[Mon May 25 05:29:33 2015] flags: 0x20047580004000(head)
[Mon May 25 05:29:33 2015] page dumped because: VM_BUG_ON_PAGE(compound_order(page) && !PageTransHuge(page))
[Mon May 25 05:29:33 2015] ------------[ cut here ]------------
[Mon May 25 05:29:33 2015] kernel BUG at mm/migrate.c:1661!
[Mon May 25 05:29:33 2015] invalid opcode: 0000 [#1] SMP
In this case it was triggered by running tcpdump but it's not necessary
reproducible on all systems.
sudo tcpdump -i bond0.100 'tcp port 4242' -c 100000000000 -w 4242.pcap
Compound pages cannot be migrated and it was not expected that such pages
be marked for NUMA balancing. This did not take into account that drivers
such as net/packet/af_packet.c may insert compound pages into userspace
with vm_insert_page. This patch tells the NUMA balancing protection
scanner to skip all VM_MIXEDMAP mappings which avoids the possibility that
compound pages are marked for migration.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Jovi Zhangwei <jovi@cloudflare.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The {producer|consumer}_{nice|fifo} parameters are integer
type, we should use 'int' as the second param in module_param.
For example(consumer_fifo):
the default value of consumer_fifo is -1.
Without this patch:
# cat /sys/module/ring_buffer_benchmark/parameters/consumer_fifo
4294967295
With this patch:
# cat /sys/module/ring_buffer_benchmark/parameters/consumer_fifo
-1
Link: http://lkml.kernel.org/r/1433923873-67712-1-git-send-email-long.wanglong@huawei.com
Signed-off-by: Wang Long <long.wanglong@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
clocksource messages aren't prefixed in dmesg so it's a bit unclear
what subsystem emits the messages.
Use pr_fmt and pr_<level> to auto-prefix the messages appropriately.
Miscellanea:
o Remove "Warning" from KERN_WARNING level messages
o Align "timekeeping watchdog: " messages
o Coalesce formats
o Align multiline arguments
Signed-off-by: Joe Perches <joe@perches.com>
Cc: John Stultz <john.stultz@linaro.org>
Link: http://lkml.kernel.org/r/1432579795.2846.75.camel@perches.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Fix the fact that @ssid is uninitialised in the case where
CGROUP_SUBSYS_COUNT = 0 by setting ssid to 0.
Fixes: cb4a316752 ("cgroup: use bitmask to filter for_each_subsys")
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Replace the explicit checking against ss_masks inside a for_each_subsys
block with for_each_subsys_which(..., ss_mask), to take advantage of the
more readable (and more efficient) macro.
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Add a new macro for_each_subsys_which that allows all enabled cgroup
subsystems to be filtered by a bitmask, such that mask & (1 << ssid)
determines if the subsystem is to be processed in the loop body (where
ssid is the unique id of the subsystem).
Also replace the need_forkexit_callback with two separate bitmasks for
each callback to make (ss->{fork,exit}) checks unnecessary.
tj: add a short comment for "if (!CGROUP_SUBSYS_COUNT)".
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
When the PEBS interrupt threshold is larger than one record and the
machine supports multiple PEBS events, the records of these events are
mixed up and we need to demultiplex them.
Demuxing the records is hard because the hardware is deficient. The
hardware has two issues that, when combined, create impossible
scenarios to demux.
The first issue is that the 'status' field of the PEBS record is a copy
of the GLOBAL_STATUS MSR at PEBS assist time. To see why this is a
problem let us first describe the regular PEBS cycle:
A) the CTRn value reaches 0:
- the corresponding bit in GLOBAL_STATUS gets set
- we start arming the hardware assist
< some unspecified amount of time later -- this could cover multiple
events of interest >
B) the hardware assist is armed, any next event will trigger it
C) a matching event happens:
- the hardware assist triggers and generates a PEBS record
this includes a copy of GLOBAL_STATUS at this moment
- if we auto-reload we (re)set CTRn
- we clear the relevant bit in GLOBAL_STATUS
Now consider the following chain of events:
A0, B0, A1, C0
The event generated for counter 0 will include a status with counter 1
set, even though its not at all related to the record. A similar thing
can happen with a !PEBS event if it just happens to overflow at the
right moment.
The second issue is that the hardware will only emit one record for two
or more counters if the event that triggers the assist is 'close'. The
'close' can be several cycles. In some cases even the complete assist,
if the event is something that doesn't need retirement.
For instance, consider this chain of events:
A0, B0, A1, B1, C01
Where C01 is an event that triggers both hardware assists, we will
generate but a single record, but again with both counters listed in the
status field.
This time the record pertains to both events.
Note that these two cases are different but undistinguishable with the
data as generated. Therefore demuxing records with multiple PEBS bits
(we can safely ignore status bits for !PEBS counters) is impossible.
Furthermore we cannot emit the record to both events because that might
cause a data leak -- the events might not have the same privileges -- so
what this patch does is discard such events.
The assumption/hope is that such discards will be rare.
Here lists some possible ways you may get high discard rate.
- when you count the same thing multiple times. But it is not a useful
configuration.
- you can be unfortunate if you measure with a userspace only PEBS
event along with either a kernel or unrestricted PEBS event. Imagine
the event triggering and setting the overflow flag right before
entering the kernel. Then all kernel side events will end up with
multiple bits set.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Kan Liang <kan.liang@intel.com>
[ Changelog improvements. ]
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: acme@infradead.org
Cc: eranian@google.com
Link: http://lkml.kernel.org/r/1430940834-8964-4-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Changeset a43455a1d5 ("sched/numa: Ensure task_numa_migrate() checks
the preferred node") fixes an issue where workloads would never
converge on a fully loaded (or overloaded) system.
However, it introduces a regression on less than fully loaded systems,
where workloads converge on a few NUMA nodes, instead of properly
staying spread out across the whole system. This leads to a reduction
in available memory bandwidth, and usable CPU cache, with predictable
performance problems.
The root cause appears to be an interaction between the load balancer
and NUMA balancing, where the short term load represented by the load
balancer differs from the long term load the NUMA balancing code would
like to base its decisions on.
Simply reverting a43455a1d5 would re-introduce the non-convergence
of workloads on fully loaded systems, so that is not a good option. As
an aside, the check done before a43455a1d5 only applied to a task's
preferred node, not to other candidate nodes in the system, so the
converge-on-too-few-nodes problem still happens, just to a lesser
degree.
Instead, try to compensate for the impedance mismatch between the load
balancer and NUMA balancing by only ever considering a lesser loaded
node as a destination for NUMA balancing, regardless of whether the
task is trying to move to the preferred node, or to another node.
This patch also addresses the issue that a system with a single
runnable thread would never migrate that thread to near its memory,
introduced by 095bebf61a ("sched/numa: Do not move past the balance
point if unbalanced").
A test where the main thread creates a large memory area, and spawns a
worker thread to iterate over the memory (placed on another node by
select_task_rq_fair), after which the main thread goes to sleep and
waits for the worker thread to loop over all the memory now sees the
worker thread migrated to where the memory is, instead of having all
the memory migrated over like before.
Jirka has run a number of performance tests on several systems: single
instance SpecJBB 2005 performance is 7-15% higher on a 4 node system,
with higher gains on systems with more cores per socket.
Multi-instance SpecJBB 2005 (one per node), linpack, and stream see
little or no changes with the revert of 095bebf61a and this patch.
Reported-by: Artem Bityutski <dedekind1@gmail.com>
Reported-by: Jirka Hladky <jhladky@redhat.com>
Tested-by: Jirka Hladky <jhladky@redhat.com>
Tested-by: Artem Bityutskiy <dedekind1@gmail.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150528095249.3083ade0@annuminas.surriel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
