linux/kernel/softirq.c, branch v3.9

Merge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2013-03-06T02:10:04Z

Pull irq fixes and cleanups from Thomas Gleixner: "Commit e5ab012c3271 ("nohz: Make tick_nohz_irq_exit() irq safe") is the first commit in the series and the minimal necessary bugfix, which needs to go back into stable. The remanining commits enforce irq disabling in irq_exit(), sanitize the hardirq/softirq preempt count transition and remove a bunch of no longer necessary conditionals." I personally love getting rid of the very subtle and confusing IRQ_EXIT_OFFSET thing. Even apart from the whole "more lines removed than added" thing. * 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: irq: Don't re-enable interrupts at the end of irq_exit irq: Remove IRQ_EXIT_OFFSET workaround Revert "nohz: Make tick_nohz_irq_exit() irq safe" irq: Sanitize invoke_softirq irq: Ensure irq_exit() code runs with interrupts disabled nohz: Make tick_nohz_irq_exit() irq safe

irq: Don't re-enable interrupts at the end of irq_exit

2013-02-28T19:00:43Z

Commit 74eed0163d0def3fce27228d9ccf3d36e207b286 "irq: Ensure irq_exit() code runs with interrupts disabled" restore interrupts flags in the end of irq_exit() for archs that don't define __ARCH_IRQ_EXIT_IRQS_DISABLED. However always returning from irq_exit() with interrupts disabled should not be a problem for these archs. Prior to this commit this was already happening anytime we processed pending softirqs anyway. Suggested-by: Linus Torvalds Signed-off-by: Frederic Weisbecker Cc: Linus Torvalds Cc: Thomas Gleixner Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Paul E. McKenney

irq: Remove IRQ_EXIT_OFFSET workaround

2013-02-21T23:05:07Z

The IRQ_EXIT_OFFSET trick was used to make sure the irq doesn't get preempted after we substract the HARDIRQ_OFFSET until we are entirely done with any code in irq_exit(). This workaround was necessary because some archs may call irq_exit() with irqs enabled and there is still some code in the end of this function that is not covered by the HARDIRQ_OFFSET but want to stay non-preemptible. Now that irq are always disabled in irq_exit(), the whole code is guaranteed not to be preempted. We can thus remove this hack. Signed-off-by: Frederic Weisbecker Cc: Linus Torvalds Cc: Thomas Gleixner Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Paul E. McKenney

irq: Sanitize invoke_softirq

2013-02-21T19:52:24Z

With the irq protection in irq_exit, we can remove the #ifdeffery and the bh_disable/enable dance in invoke_softirq() Signed-off-by: Thomas Gleixner Cc: Frederic Weisbecker Cc: Peter Zijlstra Cc: Ingo Molnar Cc: Paul E. McKenney Cc: Linus Torvalds Link: http://lkml.kernel.org/r/alpine.LFD.2.02.1302202155320.22263@ionos

irq: Ensure irq_exit() code runs with interrupts disabled

2013-02-21T19:52:24Z

We had already a few problems with code called from irq_exit() when interrupted from a nesting interrupt. This can happen on architectures which do not define __ARCH_IRQ_EXIT_IRQS_DISABLED. __ARCH_IRQ_EXIT_IRQS_DISABLED should go away and we want to make it mandatory to call irq_exit() with interrupts disabled. As a temporary protection disable interrupts for those architectures which do not define __ARCH_IRQ_EXIT_IRQS_DISABLED and add a WARN_ONCE when an architecture which defines __ARCH_IRQ_EXIT_IRQS_DISABLED calls irq_exit() with interrupts enabled. Signed-off-by: Thomas Gleixner Cc: Frederic Weisbecker Cc: Peter Zijlstra Cc: Ingo Molnar Cc: Paul E. McKenney Cc: Linus Torvalds Link: http://lkml.kernel.org/r/alpine.LFD.2.02.1302202155320.22263@ionos

Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next

2013-02-21T02:58:50Z

Pull networking update from David Miller: 1) Checkpoint/restarted TCP sockets now can properly propagate the TCP timestamp offset. From Andrey Vagin. 2) VMWARE VM VSOCK layer, from Andy King. 3) Much improved support for virtual functions and SR-IOV in bnx2x, from Ariel ELior. 4) All protocols on ipv4 and ipv6 are now network namespace aware, and all the compatability checks for initial-namespace-only protocols is removed. Thanks to Tom Parkin for helping deal with the last major holdout, L2TP. 5) IPV6 support in netpoll and network namespace support in pktgen, from Cong Wang. 6) Multiple Registration Protocol (MRP) and Multiple VLAN Registration Protocol (MVRP) support, from David Ward. 7) Compute packet lengths more accurately in the packet scheduler, from Eric Dumazet. 8) Use per-task page fragment allocator in skb_append_datato_frags(), also from Eric Dumazet. 9) Add support for connection tracking labels in netfilter, from Florian Westphal. 10) Fix default multicast group joining on ipv6, and add anti-spoofing checks to 6to4 and 6rd. From Hannes Frederic Sowa. 11) Make ipv4/ipv6 fragmentation memory limits more reasonable in modern times, rearrange inet frag datastructures for better cacheline locality, and move more operations outside of locking. From Jesper Dangaard Brouer. 12) Instead of strict master <--> slave relationships, allow arbitrary scenerios with "upper device lists". From Jiri Pirko. 13) Improve rate limiting accuracy in TBF and act_police, also from Jiri Pirko. 14) Add a BPF filter netfilter match target, from Willem de Bruijn. 15) Orphan and delete a bunch of pre-historic networking drivers from Paul Gortmaker. 16) Add TSO support for GRE tunnels, from Pravin B SHelar. Although this still needs some minor bug fixing before it's %100 correct in all cases. 17) Handle unresolved IPSEC states like ARP, with a resolution packet queue. From Steffen Klassert. 18) Remove TCP Appropriate Byte Count support (ABC), from Stephen Hemminger. This was long overdue. 19) Support SO_REUSEPORT, from Tom Herbert. 20) Allow locking a socket BPF filter, so that it cannot change after a process drops capabilities. 21) Add VLAN filtering to bridge, from Vlad Yasevich. 22) Bring ipv6 on-par with ipv4 and do not cache neighbour entries in the ipv6 routes, from YOSHIFUJI Hideaki. * git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1538 commits) ipv6: fix race condition regarding dst->expires and dst->from. net: fix a wrong assignment in skb_split() ip_gre: remove an extra dst_release() ppp: set qdisc_tx_busylock to avoid LOCKDEP splat atl1c: restore buffer state net: fix a build failure when !CONFIG_PROC_FS net: ipv4: fix waring -Wunused-variable net: proc: fix build failed when procfs is not configured Revert "xen: netback: remove redundant xenvif_put" net: move procfs code to net/core/net-procfs.c qmi_wwan, cdc-ether: add ADU960S bonding: set sysfs device_type to 'bond' bonding: fix bond_release_all inconsistencies b44: use netdev_alloc_skb_ip_align() xen: netback: remove redundant xenvif_put net: fec: Do a sanity check on the gpio number ip_gre: propogate target device GSO capability to the tunnel device ip_gre: allow CSUM capable devices to handle packets bonding: Fix initialize after use for 3ad machine state spinlock bonding: Fix race condition between bond_enslave() and bond_3ad_update_lacp_rate() ...

cputime: Safely read cputime of full dynticks CPUs

2013-01-27T19:35:47Z

While remotely reading the cputime of a task running in a full dynticks CPU, the values stored in utime/stime fields of struct task_struct may be stale. Its values may be those of the last kernel <-> user transition time snapshot and we need to add the tickless time spent since this snapshot. To fix this, flush the cputime of the dynticks CPUs on kernel <-> user transition and record the time / context where we did this. Then on top of this snapshot and the current time, perform the fixup on the reader side from task_times() accessors. Signed-off-by: Frederic Weisbecker Cc: Andrew Morton Cc: Ingo Molnar Cc: Li Zhong Cc: Namhyung Kim Cc: Paul E. McKenney Cc: Paul Gortmaker Cc: Peter Zijlstra Cc: Steven Rostedt Cc: Thomas Gleixner [fixed kvm module related build errors] Signed-off-by: Sedat Dilek

softirq: reduce latencies

2013-01-10T23:33:01Z

In various network workloads, __do_softirq() latencies can be up to 20 ms if HZ=1000, and 200 ms if HZ=100. This is because we iterate 10 times in the softirq dispatcher, and some actions can consume a lot of cycles. This patch changes the fallback to ksoftirqd condition to : - A time limit of 2 ms. - need_resched() being set on current task When one of this condition is met, we wakeup ksoftirqd for further softirq processing if we still have pending softirqs. Using need_resched() as the only condition can trigger RCU stalls, as we can keep BH disabled for too long. I ran several benchmarks and got no significant difference in throughput, but a very significant reduction of latencies (one order of magnitude) : In following bench, 200 antagonist "netperf -t TCP_RR" are started in background, using all available cpus. Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC IRQ (hard+soft) Before patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=550110.424 MIN_LATENCY=146858 MAX_LATENCY=997109 P50_LATENCY=305000 P90_LATENCY=550000 P99_LATENCY=710000 MEAN_LATENCY=376989.12 STDDEV_LATENCY=184046.92 After patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=40545.492 MIN_LATENCY=9834 MAX_LATENCY=78366 P50_LATENCY=33583 P90_LATENCY=59000 P99_LATENCY=69000 MEAN_LATENCY=38364.67 STDDEV_LATENCY=12865.26 Signed-off-by: Eric Dumazet Cc: David Miller Cc: Tom Herbert Cc: Ben Hutchings Signed-off-by: David S. Miller

cputime: Specialize irq vtime hooks

2012-10-29T20:31:32Z

With CONFIG_VIRT_CPU_ACCOUNTING, when vtime_account() is called in irq entry/exit, we perform a check on the context: if we are interrupting the idle task we account the pending cputime to idle, otherwise account to system time or its sub-areas: tsk->stime, hardirq time, softirq time, ... However this check for idle only concerns the hardirq entry and softirq entry: * Hardirq may directly interrupt the idle task, in which case we need to flush the pending CPU time to idle. * The idle task may be directly interrupted by a softirq if it calls local_bh_enable(). There is probably no such call in any idle task but we need to cover every case. Ksoftirqd is not concerned because the idle time is flushed on context switch and softirq in the end of hardirq have the idle time already flushed from the hardirq entry. In the other cases we always account to system/irq time: * On hardirq exit we account the time to hardirq time. * On softirq exit we account the time to softirq time. To optimize this and avoid the indirect call to vtime_account() and the checks it performs, specialize the vtime irq APIs and only perform the check on irq entry. Irq exit can directly call vtime_account_system(). CONFIG_IRQ_TIME_ACCOUNTING behaviour doesn't change and directly maps to its own vtime_account() implementation. One may want to take benefits from the new APIs to optimize irq time accounting as well in the future. Signed-off-by: Frederic Weisbecker Cc: Peter Zijlstra Cc: Ingo Molnar Cc: Thomas Gleixner Cc: Steven Rostedt Cc: Paul Gortmaker

Merge branch 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2012-10-01T17:43:39Z

Pull scheduler changes from Ingo Molnar: "Continued quest to clean up and enhance the cputime code by Frederic Weisbecker, in preparation for future tickless kernel features. Other than that, smallish changes." Fix up trivial conflicts due to additions next to each other in arch/{x86/}Kconfig * 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (24 commits) cputime: Make finegrained irqtime accounting generally available cputime: Gather time/stats accounting config options into a single menu ia64: Reuse system and user vtime accounting functions on task switch ia64: Consolidate user vtime accounting vtime: Consolidate system/idle context detection cputime: Use a proper subsystem naming for vtime related APIs sched: cpu_power: enable ARCH_POWER sched/nohz: Clean up select_nohz_load_balancer() sched: Fix load avg vs. cpu-hotplug sched: Remove __ARCH_WANT_INTERRUPTS_ON_CTXSW sched: Fix nohz_idle_balance() sched: Remove useless code in yield_to() sched: Add time unit suffix to sched sysctl knobs sched/debug: Limit sd->*_idx range on sysctl sched: Remove AFFINE_WAKEUPS feature flag s390: Remove leftover account_tick_vtime() header cputime: Consolidate vtime handling on context switch sched: Move cputime code to its own file cputime: Generalize CONFIG_VIRT_CPU_ACCOUNTING tile: Remove SD_PREFER_LOCAL leftover ...