Enabling the tracer selftest triggers occasionally the warning in
text_poke(), which warns when the to be modified page is not marked
reserved.
The reason is that the tracer selftest installs kprobes on functions marked
__init for testing. These probes are removed after the tests, but that
removal schedules the delayed kprobes_optimizer work, which will do the
actual text poke. If the work is executed after the init text is freed,
then the warning triggers. The bug can be reproduced reliably when the work
delay is increased.
Flush the optimizer work and wait for the optimizing/unoptimizing lists to
become empty before returning from the kprobes tracer selftest. That
ensures that all operations which were queued due to the probes removal
have completed.
Link: http://lkml.kernel.org/r/20170516094802.76a468bb@gandalf.local.home
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: stable@vger.kernel.org
Fixes: 6274de498 ("kprobes: Support delayed unoptimizing")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
The kill_css() function may be called more than once under the condition
that the css was killed but not physically removed yet followed by the
removal of the cgroup that is hosting the css. This patch prevents any
harmm from being done when that happens.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: stable@vger.kernel.org # v4.5+
This restores commit:
24b91e360e: ("nohz: Fix collision between tick and other hrtimers")
... which got reverted by commit:
558e8e27e7: ('Revert "nohz: Fix collision between tick and other hrtimers"')
... due to a regression where CPUs spuriously stopped ticking.
The bug happened when a tick fired too early past its expected expiration:
on IRQ exit the tick was scheduled again to the same deadline but skipped
reprogramming because ts->next_tick still kept in cache the deadline.
This has been fixed now with resetting ts->next_tick from the tick
itself. Extra care has also been taken to prevent from obsolete values
throughout CPU hotplug operations.
When the tick is stopped and an interrupt occurs afterward, we check on
that interrupt exit if the next tick needs to be rescheduled. If it
doesn't need any update, we don't want to do anything.
In order to check if the tick needs an update, we compare it against the
clockevent device deadline. Now that's a problem because the clockevent
device is at a lower level than the tick itself if it is implemented
on top of hrtimer.
Every hrtimer share this clockevent device. So comparing the next tick
deadline against the clockevent device deadline is wrong because the
device may be programmed for another hrtimer whose deadline collides
with the tick. As a result we may end up not reprogramming the tick
accidentally.
In a worst case scenario under full dynticks mode, the tick stops firing
as it is supposed to every 1hz, leaving /proc/stat stalled:
Task in a full dynticks CPU
----------------------------
* hrtimer A is queued 2 seconds ahead
* the tick is stopped, scheduled 1 second ahead
* tick fires 1 second later
* on tick exit, nohz schedules the tick 1 second ahead but sees
the clockevent device is already programmed to that deadline,
fooled by hrtimer A, the tick isn't rescheduled.
* hrtimer A is cancelled before its deadline
* tick never fires again until an interrupt happens...
In order to fix this, store the next tick deadline to the tick_sched
local structure and reuse that value later to check whether we need to
reprogram the clock after an interrupt.
On the other hand, ts->sleep_length still wants to know about the next
clock event and not just the tick, so we want to improve the related
comment to avoid confusion.
Reported-and-tested-by: Tim Wright <tim@binbash.co.uk>
Reported-and-tested-by: Pavel Machek <pavel@ucw.cz>
Reported-by: James Hartsock <hartsjc@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1492783255-5051-2-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
irq_set_chained_handler_and_data() sets up the chained interrupt and then
stores the handler data.
That's racy against an immediate interrupt which gets handled before the
store of the handler data happened. The handler will dereference a NULL
pointer and crash.
Cure it by storing handler data before installing the chained handler.
Reported-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
This book got converted from DocBook. Update its references to
point to the current location.
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
There are a few issues on some kernel-doc markups that was
causing troubles with kernel-doc output on ReST format:
./kernel/futex.c:492: WARNING: Inline emphasis start-string without end-string.
./kernel/futex.c:1264: WARNING: Block quote ends without a blank line; unexpected unindent.
./kernel/futex.c:1721: WARNING: Block quote ends without a blank line; unexpected unindent.
./kernel/futex.c:2338: WARNING: Block quote ends without a blank line; unexpected unindent.
./kernel/futex.c:2426: WARNING: Block quote ends without a blank line; unexpected unindent.
./kernel/futex.c:2899: WARNING: Block quote ends without a blank line; unexpected unindent.
./kernel/futex.c:2972: WARNING: Block quote ends without a blank line; unexpected unindent.
Fix them.
No functional changes.
Acked-by: Darren Hart (VMware) <dvhart@infradead.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
Pull networking fixes from David Miller:
1) Track alignment in BPF verifier so that legitimate programs won't be
rejected on !CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS architectures.
2) Make tail calls work properly in arm64 BPF JIT, from Deniel
Borkmann.
3) Make the configuration and semantics Generic XDP make more sense and
don't allow both generic XDP and a driver specific instance to be
active at the same time. Also from Daniel.
4) Don't crash on resume in xen-netfront, from Vitaly Kuznetsov.
5) Fix use-after-free in VRF driver, from Gao Feng.
6) Use netdev_alloc_skb_ip_align() to avoid unaligned IP headers in
qca_spi driver, from Stefan Wahren.
7) Always run cleanup routines in BPF samples when we get SIGTERM, from
Andy Gospodarek.
8) The mdio phy code should bring PHYs out of reset using the shared
GPIO lines before invoking bus->reset(). From Florian Fainelli.
9) Some USB descriptor access endian fixes in various drivers from
Johan Hovold.
10) Handle PAUSE advertisements properly in mlx5 driver, from Gal
Pressman.
11) Fix reversed test in mlx5e_setup_tc(), from Saeed Mahameed.
12) Cure netdev leak in AF_PACKET when using timestamping via control
messages. From Douglas Caetano dos Santos.
13) netcp doesn't support HWTSTAMP_FILTER_ALl, reject it. From Miroslav
Lichvar.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (52 commits)
ldmvsw: stop the clean timer at beginning of remove
ldmvsw: unregistering netdev before disable hardware
net: netcp: fix check of requested timestamping filter
ipv6: avoid dad-failures for addresses with NODAD
qed: Fix uninitialized data in aRFS infrastructure
mdio: mux: fix device_node_continue.cocci warnings
net/packet: fix missing net_device reference release
net/mlx4_core: Use min3 to select number of MSI-X vectors
macvlan: Fix performance issues with vlan tagged packets
net: stmmac: use correct pointer when printing normal descriptor ring
net/mlx5: Use underlay QPN from the root name space
net/mlx5e: IPoIB, Only support regular RQ for now
net/mlx5e: Fix setup TC ndo
net/mlx5e: Fix ethtool pause support and advertise reporting
net/mlx5e: Use the correct pause values for ethtool advertising
vmxnet3: ensure that adapter is in proper state during force_close
sfc: revert changes to NIC revision numbers
net: ch9200: add missing USB-descriptor endianness conversions
net: irda: irda-usb: fix firmware name on big-endian hosts
net: dsa: mv88e6xxx: add default case to switch
...
Currently, rq->leaf_cfs_rq_list is a traversal ordered list of all
live cfs_rqs which have ever been active on the CPU; unfortunately,
this makes update_blocked_averages() O(# total cgroups) which isn't
scalable at all.
This shows up as a small CPU consumption and scheduling latency
increase in the load balancing path in systems with CPU controller
enabled across most cgroups. In an edge case where temporary cgroups
were leaking, this caused the kernel to consume good several tens of
percents of CPU cycles running update_blocked_averages(), each run
taking multiple millisecs.
This patch fixes the issue by taking empty and fully decayed cfs_rqs
off the rq->leaf_cfs_rq_list.
Signed-off-by: Tejun Heo <tj@kernel.org>
[ Added cfs_rq_is_decayed() ]
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Chris Mason <clm@fb.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Turner <pjt@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20170426004350.GB3222@wtj.duckdns.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There's a discrepancy in naming between the sched_domain and
sched_group cpumask accessor. Since we're doing changes, fix it.
$ git grep sched_group_cpus | wc -l
28
$ git grep sched_domain_span | wc -l
38
Suggests changing sched_group_cpus() into sched_group_span():
for i in `git grep -l sched_group_cpus`
do
sed -ie 's/sched_group_cpus/sched_group_span/g' $i
done
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We want to attain:
sg_cpus() & sg_mask() == sg_mask()
for this to be so we must initialize sg_mask() to sg_cpus() for the
!overlap case (its currently cpumask_setall()).
Since the code makes my head hurt bad, rewrite it into a simpler form,
inspired by the now fixed overlap code.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When building the overlapping groups we need to attach a consistent
sched_group_capacity structure. That is, all 'identical' sched_group's
should have the _same_ sched_group_capacity.
This can (once again) be demonstrated with a topology like:
node 0 1 2 3
0: 10 20 30 20
1: 20 10 20 30
2: 30 20 10 20
3: 20 30 20 10
But we need at least 2 CPUs per node for this to show up, after all,
if there is only one CPU per node, our CPU @i is per definition a
unique CPU that reaches this domain (aka balance-cpu).
Given the above NUMA topo and 2 CPUs per node:
[] CPU0 attaching sched-domain(s):
[] domain-0: span=0,4 level=DIE
[] groups: 0:{ span=0 }, 4:{ span=4 }
[] domain-1: span=0-1,3-5,7 level=NUMA
[] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 }
[] domain-2: span=0-7 level=NUMA
[] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 }
[] CPU1 attaching sched-domain(s):
[] domain-0: span=1,5 level=DIE
[] groups: 1:{ span=1 }, 5:{ span=5 }
[] domain-1: span=0-2,4-6 level=NUMA
[] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 }
[] domain-2: span=0-7 level=NUMA
[] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 }
Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the
'same' but have a different id (0 vs 4).
To fix this, use the group balance CPU to select the SGC. This means
we have to compute the full mask for each CPU and require a second
temporary mask to store the group mask in (it otherwise lives in the
SGC).
The fixed topology looks like:
[] CPU0 attaching sched-domain(s):
[] domain-0: span=0,4 level=DIE
[] groups: 0:{ span=0 }, 4:{ span=4 }
[] domain-1: span=0-1,3-5,7 level=NUMA
[] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 }
[] domain-2: span=0-7 level=NUMA
[] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 }
[] CPU1 attaching sched-domain(s):
[] domain-0: span=1,5 level=DIE
[] groups: 1:{ span=1 }, 5:{ span=5 }
[] domain-1: span=0-2,4-6 level=NUMA
[] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 }
[] domain-2: span=0-7 level=NUMA
[] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 }
Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Fixes: e3589f6c81 ("sched: Allow for overlapping sched_domain spans")
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add sgc::id to easier spot domain construction issues.
Take the opportunity to slightly rework the group printing, because
adding more "(id: %d)" strings makes the entire thing very hard to
read. Also the individual groups are very hard to separate, so add
explicit visual grouping, which allows replacing all the "(%s: %d)"
format things with shorter "%s=%d" variants.
Then fix up some inconsistencies in surrounding prints for domains.
The end result looks like:
[] CPU0 attaching sched-domain(s):
[] domain-0: span=0,4 level=DIE
[] groups: 0:{ span=0 }, 4:{ span=4 }
[] domain-1: span=0-1,3-5,7 level=NUMA
[] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 }
[] domain-2: span=0-7 level=NUMA
[] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 }
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The point of sched_group_mask is to select those CPUs from
sched_group_cpus that can actually arrive at this balance domain.
The current code gets it wrong, as can be readily demonstrated with a
topology like:
node 0 1 2 3
0: 10 20 30 20
1: 20 10 20 30
2: 30 20 10 20
3: 20 30 20 10
Where (for example) domain 1 on CPU1 ends up with a mask that includes
CPU0:
[] CPU1 attaching sched-domain:
[] domain 0: span 0-2 level NUMA
[] groups: 1 (mask: 1), 2, 0
[] domain 1: span 0-3 level NUMA
[] groups: 0-2 (mask: 0-2) (cpu_capacity: 3072), 0,2-3 (cpu_capacity: 3072)
This causes sched_balance_cpu() to compute the wrong CPU and
consequently should_we_balance() will terminate early resulting in
missed load-balance opportunities.
The fixed topology looks like:
[] CPU1 attaching sched-domain:
[] domain 0: span 0-2 level NUMA
[] groups: 1 (mask: 1), 2, 0
[] domain 1: span 0-3 level NUMA
[] groups: 0-2 (mask: 1) (cpu_capacity: 3072), 0,2-3 (cpu_capacity: 3072)
(note: this relies on OVERLAP domains to always have children, this is
true because the regular topology domains are still here -- this is
before degenerate trimming)
Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Cc: stable@vger.kernel.org
Fixes: e3589f6c81 ("sched: Allow for overlapping sched_domain spans")
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We want sched_groups to be sibling child domains (or individual CPUs
when there are no child domains). Furthermore, since the first group
of a domain should include the CPU of that domain, the first group of
each domain should match the child domain.
Verify this is indeed so.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When building the overlapping groups, we very obviously should start
with the previous domain of _this_ @cpu, not CPU-0.
This can be readily demonstrated with a topology like:
node 0 1 2 3
0: 10 20 30 20
1: 20 10 20 30
2: 30 20 10 20
3: 20 30 20 10
Where (for example) CPU1 ends up generating the following nonsensical groups:
[] CPU1 attaching sched-domain:
[] domain 0: span 0-2 level NUMA
[] groups: 1 2 0
[] domain 1: span 0-3 level NUMA
[] groups: 1-3 (cpu_capacity = 3072) 0-1,3 (cpu_capacity = 3072)
Where the fact that domain 1 doesn't include a group with span 0-2 is
the obvious fail.
With patch this looks like:
[] CPU1 attaching sched-domain:
[] domain 0: span 0-2 level NUMA
[] groups: 1 0 2
[] domain 1: span 0-3 level NUMA
[] groups: 0-2 (cpu_capacity = 3072) 0,2-3 (cpu_capacity = 3072)
Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Cc: stable@vger.kernel.org
Fixes: e3589f6c81 ("sched: Allow for overlapping sched_domain spans")
Signed-off-by: Ingo Molnar <mingo@kernel.org>
With our switch to stable delayed until late_initcall(), the most
likely cause of hitting mark_tsc_unstable() is the watchdog. The
watchdog typically only triggers when creative BIOS'es fiddle with the
TSC to hide SMI latency.
Since the watchdog can only detect TSC fiddling after the fact all TSC
clocks (including userspace GTOD) can already have reported funny
values.
The only way to fully avoid this, is manually marking the TSC unstable
at boot. Suggest people do this on their broken systems.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Core2 marks its TSC unstable in ACPI Processor Idle, which is probed
after sched_init_smp(). Luckily it appears both acpi_processor and
intel_idle (which has a similar check) are mandatory built-in.
This means we can delay switching to stable until after these drivers
have ran (if they were modules, this would be impossible).
Delay the stable switch to late_initcall() to allow these drivers to
mark TSC unstable and avoid difficult stable->unstable transitions.
Reported-by: Lofstedt, Marta <marta.lofstedt@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rafael J . Wysocki <rafael.j.wysocki@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commit:
2bacec8c31 ("sched: touch softlockup watchdog after idling")
introduced the touch_softlockup_watchdog_sched() call without
justification and I feel sched_clock management is not the right
place, it should only be concerned with producing semi coherent time.
If this causes watchdog thingies, we can find a better place.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Currently we keep sched_clock_tick() active for stable TSC in order to
keep the per-CPU state semi up-to-date. The (obvious) problem is that
by the time we detect TSC is borked, our per-CPU state is also borked.
So hook into the clocksource watchdog and call a method after we've
found it to still be stable.
There's the obvious race where the TSC goes wonky between finding it
stable and us running the callback, but closing that is too much work
and not really worth it, since we're already detecting TSC wobbles
after the fact, so we cannot, per definition, fully avoid funny clock
values.
And since the watchdog runs less often than the tick, this is also an
optimization.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In the current implementation of load/util_avg, we assume that the
ongoing time segment has fully elapsed, and util/load_sum is divided
by LOAD_AVG_MAX, even if part of the time segment still remains to
run. As a consequence, this remaining part is considered as idle time
and generates unexpected variations of util_avg of a busy CPU in the
range [1002..1024[ whereas util_avg should stay at 1023.
In order to keep the metric stable, we should not consider the ongoing
time segment when computing load/util_avg but only the segments that
have already fully elapsed. But to not consider the current time
segment adds unwanted latency in the load/util_avg responsivness
especially when the time is scaled instead of the contribution.
Instead of waiting for the current time segment to have fully elapsed
before accounting it in load/util_avg, we can already account the
elapsed part but change the range used to compute load/util_avg
accordingly.
At the very beginning of a new time segment, the past segments have
been decayed and the max value is LOAD_AVG_MAX*y. At the very end of
the current time segment, the max value becomes:
LOAD_AVG_MAX*y + 1024(us) (== LOAD_AVG_MAX)
In fact, the max value is:
LOAD_AVG_MAX*y + sa->period_contrib
at any time in the time segment.
Taking advantage of the fact that:
LOAD_AVG_MAX*y == LOAD_AVG_MAX-1024
the range becomes [0..LOAD_AVG_MAX-1024+sa->period_contrib].
As the elapsed part is already accounted in load/util_sum, we update
the max value according to the current position in the time segment
instead of removing its contribution.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Morten.Rasmussen@arm.com
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: bsegall@google.com
Cc: dietmar.eggemann@arm.com
Cc: pjt@google.com
Cc: yuyang.du@intel.com
Link: http://lkml.kernel.org/r/1493188076-2767-1-git-send-email-vincent.guittot@linaro.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
I finally got around to creating trampolines for dynamically allocated
ftrace_ops with using synchronize_rcu_tasks(). For users of the ftrace
function hook callbacks, like perf, that allocate the ftrace_ops
descriptor via kmalloc() and friends, ftrace was not able to optimize
the functions being traced to use a trampoline because they would also
need to be allocated dynamically. The problem is that they cannot be
freed when CONFIG_PREEMPT is set, as there's no way to tell if a task
was preempted on the trampoline. That was before Paul McKenney
implemented synchronize_rcu_tasks() that would make sure all tasks
(except idle) have scheduled out or have entered user space.
While testing this, I triggered this bug:
BUG: unable to handle kernel paging request at ffffffffa0230077
...
RIP: 0010:0xffffffffa0230077
...
Call Trace:
schedule+0x5/0xe0
schedule_preempt_disabled+0x18/0x30
do_idle+0x172/0x220
What happened was that the idle task was preempted on the trampoline.
As synchronize_rcu_tasks() ignores the idle thread, there's nothing
that lets ftrace know that the idle task was preempted on a trampoline.
The idle task shouldn't need to ever enable preemption. The idle task
is simply a loop that calls schedule or places the cpu into idle mode.
In fact, having preemption enabled is inefficient, because it can
happen when idle is just about to call schedule anyway, which would
cause schedule to be called twice. Once for when the interrupt came in
and was returning back to normal context, and then again in the normal
path that the idle loop is running in, which would be pointless, as it
had already scheduled.
The only reason schedule_preempt_disable() enables preemption is to be
able to call sched_submit_work(), which requires preemption enabled. As
this is a nop when the task is in the RUNNING state, and idle is always
in the running state, there's no reason that idle needs to enable
preemption. But that means it cannot use schedule_preempt_disable() as
other callers of that function require calling sched_submit_work().
Adding a new function local to kernel/sched/ that allows idle to call
the scheduler without enabling preemption, fixes the
synchronize_rcu_tasks() issue, as well as removes the pointless spurious
schedule calls caused by interrupts happening in the brief window where
preemption is enabled just before it calls schedule.
Reviewed: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20170414084809.3dacde2a@gandalf.local.home
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Imagine we have a pid namespace and a task from its parent's pid_ns,
which made setns() to the pid namespace. The task is doing fork(),
while the pid namespace's child reaper is dying. We have the race
between them:
Task from parent pid_ns Child reaper
copy_process() ..
alloc_pid() ..
.. zap_pid_ns_processes()
.. disable_pid_allocation()
.. read_lock(&tasklist_lock)
.. iterate over pids in pid_ns
.. kill tasks linked to pids
.. read_unlock(&tasklist_lock)
write_lock_irq(&tasklist_lock); ..
attach_pid(p, PIDTYPE_PID); ..
.. ..
So, just created task p won't receive SIGKILL signal,
and the pid namespace will be in contradictory state.
Only manual kill will help there, but does the userspace
care about this? I suppose, the most users just inject
a task into a pid namespace and wait a SIGCHLD from it.
The patch fixes the problem. It simply checks for
(pid_ns->nr_hashed & PIDNS_HASH_ADDING) in copy_process().
We do it under the tasklist_lock, and can't skip
PIDNS_HASH_ADDING as noted by Oleg:
"zap_pid_ns_processes() does disable_pid_allocation()
and then takes tasklist_lock to kill the whole namespace.
Given that copy_process() checks PIDNS_HASH_ADDING
under write_lock(tasklist) they can't race;
if copy_process() takes this lock first, the new child will
be killed, otherwise copy_process() can't miss
the change in ->nr_hashed."
If allocation is disabled, we just return -ENOMEM
like it's made for such cases in alloc_pid().
v2: Do not move disable_pid_allocation(), do not
introduce a new variable in copy_process() and simplify
the patch as suggested by Oleg Nesterov.
Account the problem with double irq enabling
found by Eric W. Biederman.
Fixes: c876ad7682 ("pidns: Stop pid allocation when init dies")
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
CC: Andrew Morton <akpm@linux-foundation.org>
CC: Ingo Molnar <mingo@kernel.org>
CC: Peter Zijlstra <peterz@infradead.org>
CC: Oleg Nesterov <oleg@redhat.com>
CC: Mike Rapoport <rppt@linux.vnet.ibm.com>
CC: Michal Hocko <mhocko@suse.com>
CC: Andy Lutomirski <luto@kernel.org>
CC: "Eric W. Biederman" <ebiederm@xmission.com>
CC: Andrei Vagin <avagin@openvz.org>
CC: Cyrill Gorcunov <gorcunov@openvz.org>
CC: Serge Hallyn <serge@hallyn.com>
Cc: stable@vger.kernel.org
Acked-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The code can potentially sleep for an indefinite amount of time in
zap_pid_ns_processes triggering the hung task timeout, and increasing
the system average. This is undesirable. Sleep with a task state of
TASK_INTERRUPTIBLE instead of TASK_UNINTERRUPTIBLE to remove these
undesirable side effects.
Apparently under heavy load this has been allowing Chrome to trigger
the hung time task timeout error and cause ChromeOS to reboot.
Reported-by: Vovo Yang <vovoy@google.com>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Fixes: 6347e90091 ("pidns: guarantee that the pidns init will be the last pidns process reaped")
Cc: stable@vger.kernel.org
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Pull perf updates/fixes from Ingo Molnar:
"Mostly tooling updates, but also two kernel fixes: a call chain
handling robustness fix and an x86 PMU driver event definition fix"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/callchain: Force USER_DS when invoking perf_callchain_user()
tools build: Fixup sched_getcpu feature test
perf tests kmod-path: Don't fail if compressed modules aren't supported
perf annotate: Fix AArch64 comment char
perf tools: Fix spelling mistakes
perf/x86: Fix Broadwell-EP DRAM RAPL events
perf config: Refactor a duplicated code for obtaining config file name
perf symbols: Allow user probes on versioned symbols
perf symbols: Accept symbols starting at address 0
tools lib string: Adopt prefixcmp() from perf and subcmd
perf units: Move parse_tag_value() to units.[ch]
perf ui gtk: Move gtk .so name to the only place where it is used
perf tools: Move HAS_BOOL define to where perl headers are used
perf memswap: Split the byteswap memory range wrappers from util.[ch]
perf tools: Move event prototypes from util.h to event.h
perf buildid: Move prototypes from util.h to build-id.h
Pull stackprotector fixlet from Ingo Molnar:
"A single fix/enhancement to increase stackprotector canary randomness
on 64-bit kernels with very little cost"
* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
stackprotector: Increase the per-task stack canary's random range from 32 bits to 64 bits on 64-bit platforms
We must accumulate into reg->aux_off rather than use a plain assignment.
Add a test for this situation to test_align.
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a new field, "prog_flags", and an initial flag value
BPF_F_STRICT_ALIGNMENT.
When set, the verifier will enforce strict pointer alignment
regardless of the setting of CONFIG_EFFICIENT_UNALIGNED_ACCESS.
The verifier, in this mode, will also use a fixed value of "2" in
place of NET_IP_ALIGN.
This facilitates test cases that will exercise and validate this part
of the verifier even when run on architectures where alignment doesn't
matter.
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
If log_level > 1, do a state dump every instruction and emit it in
a more compact way (without a leading newline).
This will facilitate more sophisticated test cases which inspect the
verifier log for register state.
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Currently if we add only constant values to pointers we can fully
validate the alignment, and properly check if we need to reject the
program on !CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS architectures.
However, once an unknown value is introduced we only allow byte sized
memory accesses which is too restrictive.
Add logic to track the known minimum alignment of register values,
and propagate this state into registers containing pointers.
The most common paradigm that makes use of this new logic is computing
the transport header using the IP header length field. For example:
struct ethhdr *ep = skb->data;
struct iphdr *iph = (struct iphdr *) (ep + 1);
struct tcphdr *th;
...
n = iph->ihl;
th = ((void *)iph + (n * 4));
port = th->dest;
The existing code will reject the load of th->dest because it cannot
validate that the alignment is at least 2 once "n * 4" is added the
the packet pointer.
In the new code, the register holding "n * 4" will have a reg->min_align
value of 4, because any value multiplied by 4 will be at least 4 byte
aligned. (actually, the eBPF code emitted by the compiler in this case
is most likely to use a shift left by 2, but the end result is identical)
At the critical addition:
th = ((void *)iph + (n * 4));
The register holding 'th' will start with reg->off value of 14. The
pointer addition will transform that reg into something that looks like:
reg->aux_off = 14
reg->aux_off_align = 4
Next, the verifier will look at the th->dest load, and it will see
a load offset of 2, and first check:
if (reg->aux_off_align % size)
which will pass because aux_off_align is 4. reg_off will be computed:
reg_off = reg->off;
...
reg_off += reg->aux_off;
plus we have off==2, and it will thus check:
if ((NET_IP_ALIGN + reg_off + off) % size != 0)
which evaluates to:
if ((NET_IP_ALIGN + 14 + 2) % size != 0)
On strict alignment architectures, NET_IP_ALIGN is 2, thus:
if ((2 + 14 + 2) % size != 0)
which passes.
These pointer transformations and checks work regardless of whether
the constant offset or the variable with known alignment is added
first to the pointer register.
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Pull tracing fix from Steven Rostedt:
"This is a trivial patch that changes a check for a cpumask from a NULL
pointer to using cpumask_available(), which will do the check. This is
because cpumasks when not allocated are always set, and clang
complains about it"
* tag 'trace-v4.12-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
tracing: Use cpumask_available() to check if cpumask variable may be used
