I noticed expensive divides done in try_to_wakeup() and
find_busiest_group() on a bi dual core Opteron machine (total of 4 cores),
moderatly loaded (15.000 context switch per second)
oprofile numbers :
CPU: AMD64 processors, speed 2600.05 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit
mask of 0x00 (No unit mask) count 50000
samples % symbol name
...
613914 1.0498 try_to_wake_up
834 0.0013 :ffffffff80227ae1: div %rcx
77513 0.1191 :ffffffff80227ae4: mov %rax,%r11
608893 1.0413 find_busiest_group
1841 0.0031 :ffffffff802260bf: div %rdi
140109 0.2394 :ffffffff802260c2: test %sil,%sil
Some of these divides can use the reciprocal divides we introduced some
time ago (currently used in slab AFAIK)
We can assume a load will fit in a 32bits number, because with a
SCHED_LOAD_SCALE=128 value, its still a theorical limit of 33554432
When/if we reach this limit one day, probably cpus will have a fast
hardware divide and we can zap the reciprocal divide trick.
Ingo suggested to rename cpu_power to __cpu_power to make clear it should
not be modified without changing its reciprocal value too.
I did not convert the divide in cpu_avg_load_per_task(), because tracking
nr_running changes may be not worth it ? We could use a static table of 32
reciprocal values but it would add a conditional branch and table lookup.
[akpm@linux-foundation.org: !SMP build fix]
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix the process idle load balancing in the presence of dynticks. cpus for
which ticks are stopped will sleep till the next event wakes it up.
Potentially these sleeps can be for large durations and during which today,
there is no periodic idle load balancing being done.
This patch nominates an owner among the idle cpus, which does the idle load
balancing on behalf of the other idle cpus. And once all the cpus are
completely idle, then we can stop this idle load balancing too. Checks added
in fast path are minimized. Whenever there are busy cpus in the system, there
will be an owner(idle cpu) doing the system wide idle load balancing.
Open items:
1. Intelligent owner selection (like an idle core in a busy package).
2. Merge with rcu's nohz_cpu_mask?
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Periodic load balancing in recent kernels happen in the softirq. In
certain -rt configurations, these softirqs are handled in softirqd context.
And hence the check for idle processor was always returning busy (as
nr_running > 1).
This patch captures the idle information at the tick and passes this info
to softirq context through an element 'idle_at_tick' in rq.
[kernel@kolivas.org: Fix reverse idle at tick logic]
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As discovered here today, the change in Kernel 2.6.17 intended to inhibit
users from setting RLIMIT_CPU to 0 (as that is equivalent to unlimited) by
"cheating" and setting it to 1 in such a case, does not make a difference,
as the check is done in the wrong place (too late), and only applies to the
profiling code.
On all systems I checked running kernels above 2.6.17, no matter what the
hard and soft CPU time limits were before, a user could escape them by
issuing in the shell (sh/bash/zsh) "ulimit -t 0", and then the user's
process was not ever killed.
Attached is a trivial patch to fix that. Simply moving the check to a
slightly earlier location (specifically, before the line that actually
assigns the limit - *old_rlim = new_rlim), does the trick.
Do note that at least the zsh (but not ash, dash, or bash) shell has the
problem of "caching" the limits set by the ulimit command, so when running
zsh the fix will not immediately be evident - after entering "ulimit -t 0",
"ulimit -a" will show "-t: cpu time (seconds) 0", even though the actual
limit as returned by getrlimit(...) will be 1. It can be verified by
opening a subshell (which will not have the values of the parent shell in
cache) and checking in it, or just by running a CPU intensive command like
"echo '65536^1048576' | bc" and verifying that it dumps core after one
second.
Regardless of whether that is a misfeature in the shell, perhaps it would
be better to return -EINVAL from setrlimit in such a case instead of
cheating and setting to 1, as that does not really reflect the actual state
of the process anymore. I do not however know what the ground for that
decision was in the original 2.6.17 change, and whether there would be any
"backward" compatibility issues, so I preferred not to touch that right
now.
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are many places in the kernel where the construction like
foo = list_entry(head->next, struct foo_struct, list);
are used.
The code might look more descriptive and neat if using the macro
list_first_entry(head, type, member) \
list_entry((head)->next, type, member)
Here is the macro itself and the examples of its usage in the generic code.
If it will turn out to be useful, I can prepare the set of patches to
inject in into arch-specific code, drivers, networking, etc.
Signed-off-by: Pavel Emelianov <xemul@openvz.org>
Signed-off-by: Kirill Korotaev <dev@openvz.org>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: John McCutchan <ttb@tentacle.dhs.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
While the !highres/!dyntick code assigns the duty of the do_timer() call to
one specific CPU, this was dropped in the highres/dyntick part during
development.
Steven Rostedt discovered the xtime lock contention on highres/dyntick due
to several CPUs trying to update jiffies.
Add the single CPU assignement back. In the dyntick case this needs to be
handled carefully, as the CPU which has the do_timer() duty must drop the
assignement and let it be grabbed by another CPU, which is active.
Otherwise the do_timer() calls would not happen during the long sleep.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Mark Lord <mlord@pobox.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We noticed a drop in n/w performance due to the irq_desc being cacheline
aligned rather than internode aligned. We see 50% of expected performance
when two e1000 nics local to two different nodes have consecutive irq
descriptors allocated, due to false sharing.
Note that this patch does away with cacheline padding for the UP case, as
it does not seem useful for UP configurations.
Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This causes constructions like
down_write(&mm1->mmap_sem);
if (down_write_trylock(&mm2->mmap_sem)) {
...
up_write(&mm2->mmap_sem);
}
up_write(&mm1->mmap_sem);
generate a lockdep warning about circular locking dependence.
Call rwsem_acquire() with trylock set to 1.
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kallsyms_lookup() can go iterating over modules list unprotected which is OK
for emergency situations (oops), but not OK for regular stuff like
/proc/*/wchan.
Introduce lookup_symbol_name()/lookup_module_symbol_name() which copy symbol
name into caller-supplied buffer or return -ERANGE. All copying is done with
module_mutex held, so...
Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Several kallsyms_lookup() pass dummy arguments but only need, say, module's
name. Make kallsyms_lookup() accept NULLs where possible.
Also, makes picture clearer about what interfaces are needed for all symbol
resolving business.
Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
module_get_kallsym() leaks "struct module *" outside of module_mutex which is
no-no, because module can dissapear right after mutex unlock.
Copy all needed information from inside module_mutex into caller-supplied
space.
[bunk@stusta.de: is_exported() can now become static]
Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
module_get_kallsym() could in theory truncate module symbol name to fit in
buffer, but nobody does this. Always use KSYM_NAME_LEN + 1 bytes for name.
Suggested by lg^WRusty.
Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix kevent's childs priority greediness. Such tasks were always scheduled
at nice level -5 and, at that time, udev stole us the CPU time with -5.
Already posted at http://lkml.org/lkml/2005/1/10/85
[akpm@linux-foundation.org: add comment]
Signed-off-by: Jan Engelhardt <jengelh@gmx.de>
Cc: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently the size of the per-cpu region reserved to save crash notes is
set by the per-architecture value MAX_NOTE_BYTES. Which in turn is
currently set to 1024 on all supported architectures.
While testing ia64 I recently discovered that this value is in fact too
small. The particular setup I was using actually needs 1172 bytes. This
lead to very tedious failure mode where the tail of one elf note would
overwrite the head of another if they ended up being alocated sequentially
by kmalloc, which was often the case.
It seems to me that a far better approach is to caclculate the size that
the area needs to be. This patch does just that.
If a simpler stop-gap patch for ia64 to be squeezed into 2.6.21(.X) is
needed then this should be as easy as making MAX_NOTE_BYTES larger in
arch/asm-ia64/kexec.h. Perhaps 2048 would be a good choice. However, I
think that the approach in this patch is a much more robust idea.
Acked-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Remove includes of <linux/smp_lock.h> where it is not used/needed.
Suggested by Al Viro.
Builds cleanly on x86_64, i386, alpha, ia64, powerpc, sparc,
sparc64, and arm (all 59 defconfigs).
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add touch_all_softlockup_watchdogs() to allow the softlockup watchdog
timers on all cpus to be updated. This is used to prevent sysrq-t from
generating a spurious watchdog message when generating lots of output.
Softlockup watchdogs use sched_clock() as its timebase, which is inherently
per-cpu (at least, when it is measuring unstolen time). Because of this,
it isn't possible for one CPU to directly update the other CPU's timers,
but it is possible to tell the other CPUs to do update themselves
appropriately.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Acked-by: Chris Lalancette <clalance@redhat.com>
Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Cc: Rick Lindsley <ricklind@us.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Eric Dumazet
Siddha, Suresh B
Stas Sergeev
David Rientjes
Jarek Poplawski
Adrian Bunk
Tom Alsberg
Pavel Emelianov
Thomas Gleixner
Martin Peschke
Robert P. J. Day
Ravikiran G Thirumalai
Bert Wesarg
Alexey Dobriyan
Jan Engelhardt
Simon Horman
Randy Dunlap
Jeremy Fitzhardinge