system/linux
system/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux synced 2024-11-05 18:23:50 +00:00
Code Issues Projects Releases Packages Wiki Activity
linux/kernel
History
Eric Dumazet c10d73671a softirq: reduce latencies 
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 <edumazet@google.com>
Cc: David Miller <davem@davemloft.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-10 15:33:01 -08:00
..
debug KGDB/KDB fixes and cleanups 2012-10-13 11:16:58 +09:00
events Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace 2012-12-17 15:44:47 -08:00
gcov
irq irq: tsk->comm is an array 2012-12-18 15:02:11 -08:00
power Merge branch 'for-3.8' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup 2012-12-12 08:18:24 -08:00
sched sched: numa: ksm: fix oops in task_numa_placment() 2012-12-20 07:06:56 -08:00
time Merge tag 'kvm-3.8-1' of git://git.kernel.org/pub/scm/virt/kvm/kvm 2012-12-13 15:31:08 -08:00
trace Merge branch 'tip/perf/core-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace 2012-12-18 12:28:39 -08:00
.gitignore
acct.c vfs: make path_openat take a struct filename pointer 2012-10-12 20:15:09 -04:00
async.c
audit.c fs: handle failed audit_log_start properly 2012-10-09 23:33:37 -04:00
audit.h audit: optimize audit_compare_dname_path 2012-10-12 00:32:02 -04:00
audit_tree.c fsnotify: pass group to fsnotify_destroy_mark() 2012-12-11 13:44:36 -05:00
audit_watch.c Merge branch 'for-next' of git://git.infradead.org/users/eparis/notify 2012-12-20 20:11:52 -08:00
auditfilter.c audit: optimize audit_compare_dname_path 2012-10-12 00:32:02 -04:00
auditsc.c Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal 2012-12-12 12:22:13 -08:00
backtracetest.c
bounds.c
capability.c
cgroup.c Merge branch 'akpm' (Andrew's patch-bomb) 2012-12-17 20:58:12 -08:00
cgroup_freezer.c cgroup: rename ->create/post_create/pre_destroy/destroy() to ->css_alloc/online/offline/free() 2012-11-19 08:13:38 -08:00
compat.c compat: generic compat_sys_sched_rr_get_interval() implementation 2012-12-17 17:15:18 -08:00
configs.c
context_tracking.c context_tracking: New context tracking susbsystem 2012-11-30 11:40:07 -08:00
cpu.c Merge branch 'x86-bsp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip 2012-12-11 19:56:33 -08:00
cpu_pm.c
cpuset.c cpuset: use N_MEMORY instead N_HIGH_MEMORY 2012-12-12 17:38:32 -08:00
crash_dump.c
cred.c Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace 2012-12-18 10:55:28 -08:00
delayacct.c
dma.c
elfcore.c
exec_domain.c
exit.c Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace 2012-12-17 15:44:47 -08:00
extable.c
fork.c pidns: Outlaw thread creation after unshare(CLONE_NEWPID) 2012-12-24 22:53:14 -08:00
freezer.c freezer: change ptrace_stop/do_signal_stop to use freezable_schedule() 2012-10-26 14:27:49 -07:00
futex.c futex: avoid wake_futex() for a PI futex_q 2012-11-26 17:41:24 -08:00
futex_compat.c
groups.c
hrtimer.c
hung_task.c
irq_work.c
itimer.c
jump_label.c
kallsyms.c
kcmp.c kcmp: include linux/ptrace.h 2012-12-20 17:40:19 -08:00
Kconfig.freezer
Kconfig.hz
Kconfig.locks
Kconfig.preempt
kexec.c kdump: remove unneeded include 2012-10-06 03:05:19 +09:00
kfifo.c
kmod.c Bury the conditionals from kernel_thread/kernel_execve series 2012-12-19 18:07:38 -05:00
kprobes.c
ksysfs.c Merge branch 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip 2012-12-11 18:10:49 -08:00
kthread.c kthread: use N_MEMORY instead N_HIGH_MEMORY 2012-12-12 17:38:33 -08:00
latencytop.c
lglock.c
lockdep.c
lockdep_internals.h
lockdep_proc.c lockdep: Use KSYM_NAME_LEN'ed buffer for __get_key_name() 2012-10-24 12:39:09 +02:00
lockdep_states.h
Makefile Nothing all that exciting; a new module-from-fd syscall for those who want 2012-12-19 07:55:08 -08:00
modsign_certificate.S MODSIGN: Avoid using .incbin in C source 2012-12-14 13:06:44 +10:30
modsign_pubkey.c keys: use keyring_alloc() to create module signing keyring 2012-12-20 17:40:21 -08:00
module-internal.h MODSIGN: Move the magic string to the end of a module and eliminate the search 2012-10-19 17:30:40 -07:00
module.c Nothing all that exciting; a new module-from-fd syscall for those who want 2012-12-19 07:55:08 -08:00
module_signing.c MODSIGN: Don't use enum-type bitfields in module signature info block 2012-12-05 11:27:24 +10:30
mutex-debug.c
mutex-debug.h
mutex.c
mutex.h
notifier.c
nsproxy.c userns: Implement unshare of the user namespace 2012-11-20 04:18:14 -08:00
padata.c padata: use __this_cpu_read per-cpu helper 2012-12-06 17:16:23 +08:00
panic.c
params.c
pid.c pidns: Stop pid allocation when init dies 2012-12-25 16:10:05 -08:00
pid_namespace.c pidns: Stop pid allocation when init dies 2012-12-25 16:10:05 -08:00
posix-cpu-timers.c A few /dev/random improvements for the v3.8 merge window. 2012-12-19 20:23:37 -08:00
posix-timers.c
printk.c printk: boot_delay should only affect output 2012-12-17 17:15:13 -08:00
profile.c propagate name change to comments in kernel source 2012-12-06 10:39:54 +01:00
ptrace.c Merge branch 'akpm' (Andrew's patch-bomb) 2012-12-17 20:58:12 -08:00
range.c
rcu.h rcu: Add a module parameter to force use of expedited RCU primitives 2012-10-23 14:54:08 -07:00
rcupdate.c rcu: Add a module parameter to force use of expedited RCU primitives 2012-10-23 14:54:08 -07:00
rcutiny.c rcu: Fix TINY_RCU rcu_is_cpu_rrupt_from_idle check 2012-11-13 14:08:34 -08:00
rcutiny_plugin.h rcu: Add a module parameter to force use of expedited RCU primitives 2012-10-23 14:54:08 -07:00
rcutorture.c Merge branches 'urgent.2012.10.27a', 'doc.2012.11.16a', 'fixes.2012.11.13a', 'srcu.2012.10.27a', 'stall.2012.11.13a', 'tracing.2012.11.08a' and 'idle.2012.10.24a' into HEAD 2012-11-16 09:59:58 -08:00
rcutree.c context_tracking: New context tracking susbsystem 2012-11-30 11:40:07 -08:00
rcutree.h rcu: Separate accounting of callbacks from callback-free CPUs 2012-11-16 10:05:57 -08:00
rcutree_plugin.h rcu: Separate accounting of callbacks from callback-free CPUs 2012-11-16 10:05:57 -08:00
rcutree_trace.c rcu: Separate accounting of callbacks from callback-free CPUs 2012-11-16 10:05:57 -08:00
relay.c
res_counter.c res_counter: return amount of charges after res_counter_uncharge() 2012-12-18 15:02:12 -08:00
resource.c kernel/resource.c: fix stack overflow in __reserve_region_with_split() 2012-10-06 03:05:31 +09:00
rtmutex-debug.c
rtmutex-debug.h
rtmutex-tester.c
rtmutex.c
rtmutex.h
rtmutex_common.h
rwsem.c
seccomp.c seccomp: Make syscall skipping and nr changes more consistent 2012-10-02 21:14:29 +10:00
semaphore.c
signal.c Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal 2012-12-20 18:05:28 -08:00
smp.c
smpboot.c
smpboot.h
softirq.c softirq: reduce latencies 2013-01-10 15:33:01 -08:00
spinlock.c
srcu.c Merge branches 'urgent.2012.10.27a', 'doc.2012.11.16a', 'fixes.2012.11.13a', 'srcu.2012.10.27a', 'stall.2012.11.13a', 'tracing.2012.11.08a' and 'idle.2012.10.24a' into HEAD 2012-11-16 09:59:58 -08:00
stacktrace.c
stop_machine.c
sys.c cputime: Rename thread_group_times to thread_group_cputime_adjusted 2012-11-28 17:07:57 +01:00
sys_ni.c module: add syscall to load module from fd 2012-12-14 13:05:22 +10:30
sysctl.c Automatic NUMA Balancing V11 2012-12-16 15:18:08 -08:00
sysctl_binary.c pidns: Use task_active_pid_ns where appropriate 2012-11-19 05:59:09 -08:00
task_work.c
taskstats.c taskstats: cgroupstats_user_cmd() may leak on error 2012-10-06 03:05:31 +09:00
test_kprobes.c
time.c time: Move update_vsyscall definitions to timekeeper_internal.h 2012-09-24 12:38:06 -04:00
timeconst.pl
timer.c timers: Fix endless looping between cascade() and internal_add_timer() 2012-10-09 21:27:14 +02:00
tracepoint.c
tsacct.c userns: Convert taskstats to handle the user and pid namespaces. 2012-09-18 01:01:32 -07:00
uid16.c
up.c
user-return-notifier.c
user.c proc: Usable inode numbers for the namespace file descriptors. 2012-11-20 04:19:49 -08:00
user_namespace.c userns: Fix typo in description of the limitation of userns_install 2012-12-14 18:36:36 -08:00
utsname.c userns: Require CAP_SYS_ADMIN for most uses of setns. 2012-12-14 16:12:03 -08:00
utsname_sysctl.c
wait.c propagate name change to comments in kernel source 2012-12-06 10:39:54 +01:00
watchdog.c watchdog: Fix disable/enable regression 2012-12-19 12:10:33 -08:00
workqueue.c Merge branch 'for-3.8' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq 2012-12-12 08:15:13 -08:00
workqueue_sched.h