Jiri reported a machine stuck in multi_cpu_stop() with
migrate_swap_stop() as function and with the following src,dst cpu
pairs: {11, 4} {13, 11} { 4, 13}
4 11 13
cpuM: queue(4 ,13)
*Ma
cpuN: queue(13,11)
*N Na
*M Mb
cpuO: queue(11, 4)
*O Oa
*Nb
*Ob
Where *X denotes the cpu running the queueing of cpu-X and X[ab] denotes
the first/second queued work.
You'll observe the top of the workqueue for each cpu: 4,11,13 to be work
from cpus: M, O, N resp. IOW. deadlock.
Do away with the queueing trickery and introduce lg_double_lock() to
lock both CPUs and fully serialize the stop_two_cpus() callers instead
of the partial (and buggy) serialization we have now.
Reported-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150605153023.GH19282@twins.programming.kicks-ass.net
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>
The optimized task selection logic optimistically selects a new task
to run without first doing a full put_prev_task(). This is so that we
can avoid a put/set on the common ancestors of the old and new task.
Similarly, we should only call check_cfs_rq_runtime() to throttle
eligible groups if they're part of the common ancestry, otherwise it
is possible to end up with no eligible task in the simple task
selection.
Imagine:
/root
/prev /next
/A /B
If our optimistic selection ends up throttling /next, we goto simple
and our put_prev_task() ends up throttling /prev, after which we're
going to bug out in set_next_entity() because there aren't any tasks
left.
Avoid this scenario by only throttling common ancestors.
Reported-by: Mohammed Naser <mnaser@vexxhost.com>
Reported-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: Ben Segall <bsegall@google.com>
[ munged Changelog ]
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: Roman Gushchin <klamm@yandex-team.ru>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: pjt@google.com
Fixes: 678d5718d8 ("sched/fair: Optimize cgroup pick_next_task_fair()")
Link: http://lkml.kernel.org/r/xm26wq1oswoq.fsf@sword-of-the-dawn.mtv.corp.google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Since function tracing disables preemption, it needs a safe preemption
point to use when preemption is re-enabled without worrying about tracing
recursion. Ie: to avoid tracing recursion, that preemption point can't
be traced (use of notrace qualifier) and it can't call any traceable
function before that preemption point disables preemption itself, which
disarms the recursion.
preempt_schedule() was fine until commit:
b30f0e3ffe ("sched/preempt: Optimize preemption operations on __schedule() callers")
because PREEMPT_ACTIVE (which has the property to disable preemption
and this disarm tracing preemption recursion) was set before calling
any further function.
But that commit introduced the use of preempt_count_add/sub() functions
to set PREEMPT_ACTIVE and because these functions are called before
preemption gets a chance to be disabled, we have a tracing recursion.
preempt_schedule_context() is one of the possible preemption functions
used by tracing. Its special purpose is to avoid tracing recursion
against context tracking. Lets enhance this function to become more
generally tracing safe by disabling preemption with raw accessors, such
that no function is called before preemption gets disabled and disarm
the tracing recursion.
This function is going to become the specific tracing-safe preemption
point in further commit.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
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: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1433432349-1021-2-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull networking fixes from David Miller:
1) Various VTI tunnel (mark handling, PMTU) bug fixes from Alexander
Duyck and Steffen Klassert.
2) Revert ethtool PHY query change, it wasn't correct. The PHY address
selected by the driver running the PHY to MAC connection decides
what PHY address GET ethtool operations return information from.
3) Fix handling of sequence number bits for encryption IV generation in
ESP driver, from Herbert Xu.
4) UDP can return -EAGAIN when we hit a bad checksum on receive, even
when there are other packets in the receive queue which is wrong.
Just respect the error returned from the generic socket recv
datagram helper. From Eric Dumazet.
5) Fix BNA driver firmware loading on big-endian systems, from Ivan
Vecera.
6) Fix regression in that we were inheriting the congestion control of
the listening socket for new connections, the intended behavior
always was to use the default in this case. From Neal Cardwell.
7) Fix NULL deref in brcmfmac driver, from Arend van Spriel.
8) OTP parsing fix in iwlwifi from Liad Kaufman.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (26 commits)
vti6: Add pmtu handling to vti6_xmit.
Revert "net: core: 'ethtool' issue with querying phy settings"
bnx2x: Move statistics implementation into semaphores
xen: netback: read hotplug script once at start of day.
xen: netback: fix printf format string warning
Revert "netfilter: ensure number of counters is >0 in do_replace()"
net: dsa: Properly propagate errors from dsa_switch_setup_one
tcp: fix child sockets to use system default congestion control if not set
udp: fix behavior of wrong checksums
sfc: free multiple Rx buffers when required
bna: fix soft lock-up during firmware initialization failure
bna: remove unreasonable iocpf timer start
bna: fix firmware loading on big-endian machines
bridge: fix br_multicast_query_expired() bug
via-rhine: Resigning as maintainer
brcmfmac: avoid null pointer access when brcmf_msgbuf_get_pktid() fails
mac80211: Fix mac80211.h docbook comments
iwlwifi: nvm: fix otp parsing in 8000 hw family
iwlwifi: pcie: fix tracking of cmd_in_flight
ip_vti/ip6_vti: Preserve skb->mark after rcv_cb call
...
Pull Sparc fixes from David Miller:
1) Setup the core/threads/sockets bitmaps correctly so that 'lscpus'
and friends operate properly. Frtom Chris Hyser.
2) The bit that normally means "Cached Virtually" on sun4v systems,
actually changes meaning in M7 and later chips. Fix from Khalid
Aziz.
3) One some PCI-E systems we need to probe different OF properties to
fill in the PCI slot information properly, from Eric Snowberg.
4) Kill an extraneous memset after kzalloc(), from Christophe Jaillet.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc:
sparc: Resolve conflict between sparc v9 and M7 on usage of bit 9 of TTE
sparc64: pci slots information is not populated in sysfs
sparc: kernel: GRPCI2: Remove a useless memset
sparc64: Setup sysfs to mark LDOM sockets, cores and threads correctly
Pull virtio fix from Michael Tsirkin:
"Last-minute virtio fix for 4.1
This tweaks an exported user-space header to fix build breakage for
userspace using it"
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
include/uapi/linux/virtio_balloon.h: include linux/virtio_types.h
Pablo Neira Ayuso says:
====================
Netfilter fix for net
The following patch reverts the ebtables chunk that enforces counters that was
introduced in the recently applied d26e2c9ffa ('Revert "netfilter: ensure
number of counters is >0 in do_replace()"') since this breaks ebtables.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Kalle Valo says:
====================
iwlwifi:
* fix OTP parsing 8260
* fix powersave handling for 8260
brcmfmac:
* fix null pointer crash
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
We currently rely on the PMTU discovery of xfrm.
However if a packet is localy sent, the PMTU mechanism
of xfrm tries to to local socket notification what
might not work for applications like ping that don't
check for this. So add pmtu handling to vti6_xmit to
report MTU changes immediately.
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit f96dee13b8.
It isn't right, ethtool is meant to manage one PHY instance
per netdevice at a time, and this is selected by the SET
command. Therefore by definition the GET command must only
return the settings for the configured and selected PHY.
Reported-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit dff173de84 ("bnx2x: Fix statistics locking scheme") changed the
bnx2x locking around statistics state into using a mutex - but the lock
is being accessed via a timer which is forbidden.
[If compiled with CONFIG_DEBUG_MUTEXES, logs show a warning about
accessing the mutex in interrupt context]
This moves the implementation into using a semaphore [with size '1']
instead.
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: Ariel Elior <Ariel.Elior@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we come to tear things down in netback_remove() and generate the
uevent it is possible that the xenstore directory has already been
removed (details below).
In such cases netback_uevent() won't be able to read the hotplug
script and will write a xenstore error node.
A recent change to the hypervisor exposed this race such that we now
sometimes lose it (where apparently we didn't ever before).
Instead read the hotplug script configuration during setup and use it
for the lifetime of the backend device.
The apparently more obvious fix of moving the transition to
state=Closed in netback_remove() to after the uevent does not work
because it is possible that we are already in state=Closed (in
reaction to the guest having disconnected as it shutdown). Being
already in Closed means the toolstack is at liberty to start tearing
down the xenstore directories. In principal it might be possible to
arrange to unregister the device sooner (e.g on transition to Closing)
such that xenstore would still be there but this state machine is
fragile and prone to anger...
A modern Xen system only relies on the hotplug uevent for driver
domains, when the backend is in the same domain as the toolstack it
will run the necessary setup/teardown directly in the correct sequence
wrt xenstore changes.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/xen-netback/netback.c: In function ‘xenvif_tx_build_gops’:
drivers/net/xen-netback/netback.c:1253:8: warning: format ‘%lu’ expects argument of type ‘long unsigned int’, but argument 5 has type ‘int’ [-Wformat=]
(txreq.offset&~PAGE_MASK) + txreq.size);
^
PAGE_MASK's type can vary by arch, so a cast is needed.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
----
v2: Cast to unsigned long, since PAGE_MASK can vary by arch.
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Wanpeng Li
Peter Zijlstra
Mathieu Desnoyers
Srikar Dronamraju
Rik van Riel
Ben Segall
Frederic Weisbecker
Ingo Molnar
Linus Torvalds
David S. Miller
Steffen Klassert
Yuval Mintz
Ian Campbell