Impact: kernel crash fix
Yanmin Zhang reported that using a PERF_COUNT_TASK_CLOCK software
counter as a per-cpu counter would reliably crash the system, because
it calls __task_delta_exec with a null pointer. The page fault,
context switch and cpu migration counters also won't function
correctly as per-cpu counters since they reference the current task.
This fixes the problem by redirecting the task_clock counter to the
cpu_clock counter when used as a per-cpu counter, and by implementing
per-cpu page fault, context switch and cpu migration counters.
Along the way, this:
- Initializes counter->ctx earlier, in perf_counter_alloc, so that
sw_perf_counter_init can use it
- Adds code to kernel/sched.c to count task migrations into each
cpu, in rq->nr_migrations_in
- Exports the per-cpu context switch and task migration counts
via new functions added to kernel/sched.c
- Makes sure that if sw_perf_counter_init fails, we don't try to
initialize the counter as a hardware counter. Since the user has
passed a negative, non-raw event type, they clearly don't intend
for it to be interpreted as a hardware event.
Reported-by: "Zhang Yanmin" <yanmin_zhang@linux.intel.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
With exclusive waiters, every process woken up through the wait queue must
ensure that the next waiter down the line is woken when it has finished.
Interruptible waiters don't do that when aborting due to a signal. And if
an aborting waiter is concurrently woken up through the waitqueue, noone
will ever wake up the next waiter.
This has been observed with __wait_on_bit_lock() used by
lock_page_killable(): the first contender on the queue was aborting when
the actual lock holder woke it up concurrently. The aborted contender
didn't acquire the lock and therefor never did an unlock followed by
waking up the next waiter.
Add abort_exclusive_wait() which removes the process' wait descriptor from
the waitqueue, iff still queued, or wakes up the next waiter otherwise.
It does so under the waitqueue lock. Racing with a wake up means the
aborting process is either already woken (removed from the queue) and will
wake up the next waiter, or it will remove itself from the queue and the
concurrent wake up will apply to the next waiter after it.
Use abort_exclusive_wait() in __wait_event_interruptible_exclusive() and
__wait_on_bit_lock() when they were interrupted by other means than a wake
up through the queue.
[akpm@linux-foundation.org: coding-style fixes]
Reported-by: Chris Mason <chris.mason@oracle.com>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Mentored-by: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Chuck Lever <cel@citi.umich.edu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: <stable@kernel.org> ["after some testing"]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Christian Borntraeger reports:
> After a logical cpu offline, even on a complete idle system, there
> is one cpu with full ticks. It turns out that nohz.cpu_mask has the
> the offlined cpu still set.
>
> In select_nohz_load_balancer() we check if the system is completely
> idle to turn of load balancing. We compare cpu_online_map with
> nohz.cpu_mask. Since cpu_online_map is updated on cpu unplug,
> but nohz.cpu_mask is not, the check fails and the scheduler believes
> that we need an "idle load balancer" even on a fully idle system.
> Since the ilb cpu does not deactivate the timer tick this breaks NOHZ.
Fix the select_nohz_load_balancer() to not set the nohz.cpu_mask
while a cpu is going offline.
Reported-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Reinstate the weakening of the sync hint if set. This yields a more
symmetric usage of avg_overlap.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Pawel Dziekonski reported that the openssl benchmark and his
quantum chemistry application both show slowdowns due to the
scheduler under-parallelizing execution.
The reason are pipe wakeups still doing 'sync' wakeups which
overrides the normal buddy wakeup logic - even if waker and
wakee are loosely coupled.
Fix an inversion of logic in the buddy wakeup code.
Reported-by: Pawel Dziekonski <dzieko@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix to preempt trace triggering lockdep check_flag failure
In local_bh_disable, the use of add_preempt_count causes the
preempt tracer to start recording the time preemption is off.
But because it already modified the preempt_count to show
softirqs disabled, and before it called the lockdep code to
handle this, it causes a state that lockdep can not handle.
The preempt tracer will reset the ring buffer on start of a trace,
and the ring buffer reset code does a spin_lock_irqsave. This
calls into lockdep and lockdep will fail when it detects the
invalid state of having softirqs disabled but the internal
current->softirqs_enabled is still set.
The fix is to manually add the SOFTIRQ_OFFSET to preempt count
and call the preempt tracer code outside the lockdep critical
area.
Thanks to Peter Zijlstra for suggesting this solution.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Conflicts:
arch/x86/include/asm/pda.h
We merge tip/core/percpu into tip/perfcounters/core because of a
semantic and contextual conflict: the former eliminates the PDA,
while the latter extends it with apic_perf_irqs field.
Resolve the conflict by moving the new field to the irq_cpustat
structure on 64-bit too.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Increase the SCHED_IDLE weight from 2 to 3, this gives much more stable
vruntime numbers.
time advanced in 100ms:
weight=2
64765.988352
67012.881408
88501.412352
weight=3
35496.181411
34130.971298
35497.411573
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: make rt-limit tunables work again
Mark Glines reported:
> I've got an issue on x86-64 where I can't configure the system to allow
> RT tasks for a non-root user.
>
> In 2.6.26.5, I was able to do the following to set things up nicely:
> echo 450000 >/sys/kernel/uids/0/cpu_rt_runtime
> echo 450000 >/sys/kernel/uids/1000/cpu_rt_runtime
>
> Seems like every value I try to echo into the /sys files returns EINVAL.
For UID grouping we initialize the root group with infinite bandwidth
which by default is actually more than the global limit, therefore the
bandwidth check always fails.
Because the root group is a phantom group (for UID grouping) we cannot
runtime adjust it, therefore we let it reflect the global bandwidth
settings.
Reported-by: Mark Glines <mark@glines.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Introduce a new avg_wakeup statistic.
avg_wakeup is a measure of how frequently a task wakes up other tasks, it
represents the average time between wakeups, with a limit of avg_runtime
for when it doesn't wake up anybody.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Change mutex contention behaviour such that it will sometimes busy wait on
acquisition - moving its behaviour closer to that of spinlocks.
This concept got ported to mainline from the -rt tree, where it was originally
implemented for rtmutexes by Steven Rostedt, based on work by Gregory Haskins.
Testing with Ingo's test-mutex application (http://lkml.org/lkml/2006/1/8/50)
gave a 345% boost for VFS scalability on my testbox:
# ./test-mutex-shm V 16 10 | grep "^avg ops"
avg ops/sec: 296604
# ./test-mutex-shm V 16 10 | grep "^avg ops"
avg ops/sec: 85870
The key criteria for the busy wait is that the lock owner has to be running on
a (different) cpu. The idea is that as long as the owner is running, there is a
fair chance it'll release the lock soon, and thus we'll be better off spinning
instead of blocking/scheduling.
Since regular mutexes (as opposed to rtmutexes) do not atomically track the
owner, we add the owner in a non-atomic fashion and deal with the races in
the slowpath.
Furthermore, to ease the testing of the performance impact of this new code,
there is means to disable this behaviour runtime (without having to reboot
the system), when scheduler debugging is enabled (CONFIG_SCHED_DEBUG=y),
by issuing the following command:
# echo NO_OWNER_SPIN > /debug/sched_features
This command re-enables spinning again (this is also the default):
# echo OWNER_SPIN > /debug/sched_features
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The problem is that dropping the spinlock right before schedule is a voluntary
preemption point and can cause a schedule, right after which we schedule again.
Fix this inefficiency by keeping preemption disabled until we schedule, do this
by explicity disabling preemption and providing a schedule() variant that
assumes preemption is already disabled.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Ingo found a build error in the scheduler when RT_GROUP_SCHED was
enabled, but SMP was not. This patch rearranges the code such
that it is a little more streamlined and compiles under all permutations
of SMP, UP and RT_GROUP_SCHED. It was boot tested on my 4-way x86_64
and it still passes preempt-test.
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
This reverts commit 7317d7b87e.
This has been reported (and bisected) by Alexey Zaytsev and
Kamalesh Babulal to produce annoying warnings during bootup
on both x86 and powerpc.
kernel_locked() is not a valid test in IRQ context (we update the
BKL's ->lock_depth and the preempt count separately and non-atomicalyy),
so we cannot put it into the generic preempt debugging checks which
can run in IRQ contexts too.
Reported-and-bisected-by: Alexey Zaytsev <alexey.zaytsev@gmail.com>
Reported-and-bisected-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: build fix on certain configs
Added 'double_rq_lock' forward declaration, allowing double_rq_lock
to be used in _double_lock_balance().
Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix panic on ia64 with NR_CPUS=1024
struct sched_domain is now a dangling structure; where we really want
static ones, we need to use static_sched_domain.
(As the FIXME in this file says, cpumask_var_t would be better, but
this code is hairy enough without trying to add initialization code to
the right places).
Reported-by: Mike Travis <travis@sgi.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Vaidyanathan Srinivasan reported:
> =============================================
> [ INFO: possible recursive locking detected ]
> 2.6.28-autotest-tip-sv #1
> ---------------------------------------------
> klogd/5062 is trying to acquire lock:
> (&rq->lock){++..}, at: [<ffffffff8022aca2>] task_rq_lock+0x45/0x7e
>
> but task is already holding lock:
> (&rq->lock){++..}, at: [<ffffffff805f7354>] schedule+0x158/0xa31
With sched_mc at 2. (it is default-off)
Strictly speaking we'll not deadlock, because ttwu will not be able to
place the migration task on our rq, but since the code can deal with
both rqs getting unlocked, this seems the easiest way out.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
init_rootdomain() calls alloc_bootmem_cpumask_var() at system boot,
so does cpupri_init().
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
It's not the responsibility of init_rootdomain() to free root_domain
allocated by alloc_rootdomain().
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Reviewed-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
- Make arch_reinit_sched_domains() static. It was exported to be used in
s390, but now rebuild_sched_domains() is used instead.
- Make it return void.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
The only caller is cpu_dev_init() which is marked as __init.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'cpus4096-for-linus-3' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (77 commits)
x86: setup_per_cpu_areas() cleanup
cpumask: fix compile error when CONFIG_NR_CPUS is not defined
cpumask: use alloc_cpumask_var_node where appropriate
cpumask: convert shared_cpu_map in acpi_processor* structs to cpumask_var_t
x86: use cpumask_var_t in acpi/boot.c
x86: cleanup some remaining usages of NR_CPUS where s/b nr_cpu_ids
sched: put back some stack hog changes that were undone in kernel/sched.c
x86: enable cpus display of kernel_max and offlined cpus
ia64: cpumask fix for is_affinity_mask_valid()
cpumask: convert RCU implementations, fix
xtensa: define __fls
mn10300: define __fls
m32r: define __fls
h8300: define __fls
frv: define __fls
cris: define __fls
cpumask: CONFIG_DISABLE_OBSOLETE_CPUMASK_FUNCTIONS
cpumask: zero extra bits in alloc_cpumask_var_node
cpumask: replace for_each_cpu_mask_nr with for_each_cpu in kernel/time/
cpumask: convert mm/
...
Impact: prevents panic from stack overflow on numa-capable machines.
Some of the "removal of stack hogs" changes in kernel/sched.c by using
node_to_cpumask_ptr were undone by the early cpumask API updates, and
causes a panic due to stack overflow. This patch undoes those changes
by using cpumask_of_node() which returns a 'const struct cpumask *'.
In addition, cpu_coregoup_map is replaced with cpu_coregroup_mask further
reducing stack usage. (Both of these updates removed 9 FIXME's!)
Also:
Pick up some remaining changes from the old 'cpumask_t' functions to
the new 'struct cpumask *' functions.
Optimize memory traffic by allocating each percpu local_cpu_mask on the
same node as the referring cpu.
Signed-off-by: Mike Travis <travis@sgi.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'cpus4096-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (66 commits)
x86: export vector_used_by_percpu_irq
x86: use logical apicid in x2apic_cluster's x2apic_cpu_mask_to_apicid_and()
sched: nominate preferred wakeup cpu, fix
x86: fix lguest used_vectors breakage, -v2
x86: fix warning in arch/x86/kernel/io_apic.c
sched: fix warning in kernel/sched.c
sched: move test_sd_parent() to an SMP section of sched.h
sched: add SD_BALANCE_NEWIDLE at MC and CPU level for sched_mc>0
sched: activate active load balancing in new idle cpus
sched: bias task wakeups to preferred semi-idle packages
sched: nominate preferred wakeup cpu
sched: favour lower logical cpu number for sched_mc balance
sched: framework for sched_mc/smt_power_savings=N
sched: convert BALANCE_FOR_xx_POWER to inline functions
x86: use possible_cpus=NUM to extend the possible cpus allowed
x86: fix cpu_mask_to_apicid_and to include cpu_online_mask
x86: update io_apic.c to the new cpumask code
x86: Introduce topology_core_cpumask()/topology_thread_cpumask()
x86: xen: use smp_call_function_many()
x86: use work_on_cpu in x86/kernel/cpu/mcheck/mce_amd_64.c
...
Fixed up trivial conflict in kernel/time/tick-sched.c manually
The cpu time spent by the idle process actually doing something is
currently accounted as idle time. This is plain wrong, the architectures
that support VIRT_CPU_ACCOUNTING=y can do better: distinguish between the
time spent doing nothing and the time spent by idle doing work. The first
is accounted with account_idle_time and the second with account_system_time.
The architectures that use the account_xxx_time interface directly and not
the account_xxx_ticks interface now need to do the check for the idle
process in their arch code. In particular to improve the system vs true
idle time accounting the arch code needs to measure the true idle time
instead of just testing for the idle process.
To improve the tick based accounting as well we would need an architecture
primitive that can tell us if the pt_regs of the interrupted context
points to the magic instruction that halts the cpu.
In addition idle time is no more added to the stime of the idle process.
This field now contains the system time of the idle process as it should
be. On systems without VIRT_CPU_ACCOUNTING this will always be zero as
every tick that occurs while idle is running will be accounted as idle
time.
This patch contains the necessary common code changes to be able to
distinguish idle system time and true idle time. The architectures with
support for VIRT_CPU_ACCOUNTING need some changes to exploit this.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The utimescaled / stimescaled fields in the task structure and the
global cpustat should be set on all architectures. On s390 the calls
to account_user_time_scaled and account_system_time_scaled never have
been added. In addition system time that is accounted as guest time
to the user time of a process is accounted to the scaled system time
instead of the scaled user time.
To fix the bugs and to prevent future forgetfulness this patch merges
account_system_time_scaled into account_system_time and
account_user_time_scaled into account_user_time.
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Michael Neuling <mikey@neuling.org>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
* 'timers-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
hrtimers: fix warning in kernel/hrtimer.c
x86: make sure we really have an hpet mapping before using it
x86: enable HPET on Fujitsu u9200
linux/timex.h: cleanup for userspace
posix-timers: simplify de_thread()->exit_itimers() path
posix-timers: check ->it_signal instead of ->it_pid to validate the timer
posix-timers: use "struct pid*" instead of "struct task_struct*"
nohz: suppress needless timer reprogramming
clocksource, acpi_pm.c: put acpi_pm_read_slow() under CONFIG_PCI
nohz: no softirq pending warnings for offline cpus
hrtimer: removing all ur callback modes, fix
hrtimer: removing all ur callback modes, fix hotplug
hrtimer: removing all ur callback modes
x86: correct link to HPET timer specification
rtc-cmos: export second NVRAM bank
Fixed up conflicts in sound/drivers/pcsp/pcsp.c and sound/core/hrtimer.c
manually.
* 'core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (63 commits)
stacktrace: provide save_stack_trace_tsk() weak alias
rcu: provide RCU options on non-preempt architectures too
printk: fix discarding message when recursion_bug
futex: clean up futex_(un)lock_pi fault handling
"Tree RCU": scalable classic RCU implementation
futex: rename field in futex_q to clarify single waiter semantics
x86/swiotlb: add default swiotlb_arch_range_needs_mapping
x86/swiotlb: add default phys<->bus conversion
x86: unify pci iommu setup and allow swiotlb to compile for 32 bit
x86: add swiotlb allocation functions
swiotlb: consolidate swiotlb info message printing
swiotlb: support bouncing of HighMem pages
swiotlb: factor out copy to/from device
swiotlb: add arch hook to force mapping
swiotlb: allow architectures to override phys<->bus<->phys conversions
swiotlb: add comment where we handle the overflow of a dma mask on 32 bit
rcu: fix rcutorture behavior during reboot
resources: skip sanity check of busy resources
swiotlb: move some definitions to header
swiotlb: allow architectures to override swiotlb pool allocation
...
Fix up trivial conflicts in
arch/x86/kernel/Makefile
arch/x86/mm/init_32.c
include/linux/hardirq.h
as per Ingo's suggestions.
The RT scheduler employs a "push/pull" design to actively balance tasks
within the system (on a per disjoint cpuset basis). When a task is
awoken, it is immediately determined if there are any lower priority
cpus which should be preempted. This is opposed to the way normal
SCHED_OTHER tasks behave, which will wait for a periodic rebalancing
operation to occur before spreading out load.
When a particular RQ has more than 1 active RT task, it is said to
be in an "overloaded" state. Once this occurs, the system enters
the active balancing mode, where it will try to push the task away,
or persuade a different cpu to pull it over. The system will stay
in this state until the system falls back below the <= 1 queued RT
task per RQ.
However, the current implementation suffers from a limitation in the
push logic. Once overloaded, all tasks (other than current) on the
RQ are analyzed on every push operation, even if it was previously
unpushable (due to affinity, etc). Whats more, the operation stops
at the first task that is unpushable and will not look at items
lower in the queue. This causes two problems:
1) We can have the same tasks analyzed over and over again during each
push, which extends out the fast path in the scheduler for no
gain. Consider a RQ that has dozens of tasks that are bound to a
core. Each one of those tasks will be encountered and skipped
for each push operation while they are queued.
2) There may be lower-priority tasks under the unpushable task that
could have been successfully pushed, but will never be considered
until either the unpushable task is cleared, or a pull operation
succeeds. The net result is a potential latency source for mid
priority tasks.
This patch aims to rectify these two conditions by introducing a new
priority sorted list: "pushable_tasks". A task is added to the list
each time a task is activated or preempted. It is removed from the
list any time it is deactivated, made current, or fails to push.
This works because a task only needs to be attempted to push once.
After an initial failure to push, the other cpus will eventually try to
pull the task when the conditions are proper. This also solves the
problem that we don't completely analyze all tasks due to encountering
an unpushable tasks. Now every task will have a push attempted (when
appropriate).
This reduces latency both by shorting the critical section of the
rq->lock for certain workloads, and by making sure the algorithm
considers all eligible tasks in the system.
[ rostedt: added a couple more BUG_ONs ]
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Steven Rostedt <srostedt@redhat.com>
We currently run class->post_schedule() outside of the rq->lock, which
means that we need to test for the need to post_schedule outside of
the lock to avoid a forced reacquistion. This is currently not a problem
as we only look at rq->rt.overloaded. However, we want to enhance this
going forward to look at more state to reduce the need to post_schedule to
a bare minimum set. Therefore, we introduce a new member-func called
needs_post_schedule() which tests for the post_schedule condtion without
actually performing the work. Therefore it is safe to call this
function before the rq->lock is released, because we are guaranteed not
to drop the lock at an intermediate point (such as what post_schedule()
may do).
We will use this later in the series
[ rostedt: removed paranoid BUG_ON ]
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
double_lock balance() currently favors logically lower cpus since they
often do not have to release their own lock to acquire a second lock.
The result is that logically higher cpus can get starved when there is
a lot of pressure on the RQs. This can result in higher latencies on
higher cpu-ids.
This patch makes the algorithm more fair by forcing all paths to have
to release both locks before acquiring them again. Since callsites to
double_lock_balance already consider it a potential preemption/reschedule
point, they have the proper logic to recheck for atomicity violations.
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
git-id c4acb2c066 attempted to limit
newidle critical section length by stopping after at least one task
was moved. Further investigation has shown that there are other
paths nested further inside the algorithm which still remain that allow
long latencies to occur with newidle balancing. This patch applies
the same technique inside balance_tasks() to limit the duration of
this optional balancing operation.
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
CC: Nick Piggin <npiggin@suse.de>
We will use this later in the series to reduce the amount of rq-lock
contention during a pull operation
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
* 'tracing-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (241 commits)
sched, trace: update trace_sched_wakeup()
tracing/ftrace: don't trace on early stage of a secondary cpu boot, v3
Revert "x86: disable X86_PTRACE_BTS"
ring-buffer: prevent false positive warning
ring-buffer: fix dangling commit race
ftrace: enable format arguments checking
x86, bts: memory accounting
x86, bts: add fork and exit handling
ftrace: introduce tracing_reset_online_cpus() helper
tracing: fix warnings in kernel/trace/trace_sched_switch.c
tracing: fix warning in kernel/trace/trace.c
tracing/ring-buffer: remove unused ring_buffer size
trace: fix task state printout
ftrace: add not to regex on filtering functions
trace: better use of stack_trace_enabled for boot up code
trace: add a way to enable or disable the stack tracer
x86: entry_64 - introduce FTRACE_ frame macro v2
tracing/ftrace: add the printk-msg-only option
tracing/ftrace: use preempt_enable_no_resched_notrace in ring_buffer_time_stamp()
x86, bts: correctly report invalid bts records
...
Fixed up trivial conflict in scripts/recordmcount.pl due to SH bits
being already partly merged by the SH merge.
cpu_coregroup_map returned a cpumask_t: it's going away.
(Note, the sched part of this patch won't apply meaningfully to the
sched tree, but I'm posting it to show the goal).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Ingo Molnar <mingo@redhat.com>
Impact: extend the wakeup tracepoint with the info whether the wakeup was real
Add the information needed to distinguish 'real' wakeups from 'false'
wakeups.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Andrew Morton reported:
> kernel/sched.c: In function 'schedule':
> kernel/sched.c:3679: warning: 'active_balance' may be used uninitialized in this function
>
> This warning is correct - the code is buggy.
In sched.c load_balance_newidle, there's real potential use of
uninitialised variable - fix it.
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix cpumask conversion bug
this warning:
kernel/sched.c: In function ‘find_busiest_group’:
kernel/sched.c:3429: warning: passing argument 1 of ‘__first_cpu’ from incompatible pointer type
shows that we forgot to convert a new patch to the new cpumask APIs.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: tweak task balancing to save power more agressively
Active load balancing is a process by which migration thread
is woken up on the target CPU in order to pull current
running task on another package into this newly idle
package.
This method is already in use with normal load_balance(),
this patch introduces this method to new idle cpus when
sched_mc is set to POWERSAVINGS_BALANCE_WAKEUP.
This logic provides effective consolidation of short running
daemon jobs in a almost idle system
The side effect of this patch may be ping-ponging of tasks
if the system is moderately utilised. May need to adjust the
iterations before triggering.
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: extend load-balancing code (no change in behavior yet)
When the system utilisation is low and more cpus are idle,
then the process waking up from sleep should prefer to
wakeup an idle cpu from semi-idle cpu package (multi core
package) rather than a completely idle cpu package which
would waste power.
Use the sched_mc balance logic in find_busiest_group() to
nominate a preferred wakeup cpu.
This info can be stored in appropriate sched_domain, but
updating this info in all copies of sched_domain is not
practical. Hence this information is stored in root_domain
struct which is one copy per partitioned sched domain.
The root_domain can be accessed from each cpu's runqueue
and there is one copy per partitioned sched domain.
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: change load-balancing direction to match that of irqbalanced
Just in case two groups have identical load, prefer to move load to lower
logical cpu number rather than the present logic of moving to higher logical
number.
find_busiest_group() tries to look for a group_leader that has spare capacity
to take more tasks and freeup an appropriate least loaded group. Just in case
there is a tie and the load is equal, then the group with higher logical number
is favoured. This conflicts with user space irqbalance daemon that will move
interrupts to lower logical number if the system utilisation is very low.
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: extend range of /sys/devices/system/cpu/sched_mc_power_savings
Currently the sched_mc/smt_power_savings variable is a boolean,
which either enables or disables topology based power savings.
This patch extends the behaviour of the variable from boolean to
multivalued, such that based on the value, we decide how
aggressively do we want to perform powersavings balance at
appropriate sched domain based on topology.
Variable levels of power saving tunable would benefit end user to
match the required level of power savings vs performance
trade-off depending on the system configuration and workloads.
This version makes the sched_mc_power_savings global variable to
take more values (0,1,2). Later versions can have a single
tunable called sched_power_savings instead of
sched_{mc,smt}_power_savings.
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
these warnings:
kernel/trace/trace_sched_switch.c: In function ‘tracing_sched_register’:
kernel/trace/trace_sched_switch.c:96: warning: passing argument 1 of ‘register_trace_sched_wakeup_new’ from incompatible pointer type
kernel/trace/trace_sched_switch.c:112: warning: passing argument 1 of ‘unregister_trace_sched_wakeup_new’ from incompatible pointer type
kernel/trace/trace_sched_switch.c: In function ‘tracing_sched_unregister’:
kernel/trace/trace_sched_switch.c:121: warning: passing argument 1 of ‘unregister_trace_sched_wakeup_new’ from incompatible pointer type
Trigger because sched_wakeup_new tracepoints need the same trace
signature as sched_wakeup - which was changed recently.
Fix it.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: simplify code
When we turn on CONFIG_SCHEDSTATS, per-task cpu runtime is accumulated
twice. Once in task->se.sum_exec_runtime and once in sched_info.cpu_time.
These two stats are exactly the same.
Given that task->se.sum_exec_runtime is always accumulated by the core
scheduler, sched_info can reuse that data instead of duplicate the accounting.
Signed-off-by: Ken Chen <kenchen@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch export per-cpu CPU cycle usage for a given cpuacct cgroup.
There is a need for a user space monitor daemon to track group CPU
usage on per-cpu base. It is also useful for monitoring CFS load
balancer behavior by tracking per CPU group usage.
Signed-off-by: Ken Chen <kenchen@google.com>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: micro-optimize the code on 64-bit architectures
In the thread regarding to 'export percpu cpuacct cgroup stats'
http://lkml.org/lkml/2008/12/7/13
akpm pointed out that current cpuacct code is inefficient. This patch
refactoring the following:
* make cpu_rq locking only on 32-bit
* change iterator to each_present_cpu instead of each_possible_cpu to
make it hotplug friendly.
It's a bit of code churn, but I was rewarded with 160 byte code size saving
on x86-64 arch and zero code size change on i386.
Signed-off-by: Ken Chen <kenchen@google.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: sharpen the wakeup-granularity to always be against current scheduler time
It was possible to do the preemption check against an old time stamp.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: add new feature, new sw counter
Add a counter that counts the number of cross-CPU migrations a
task is suffering.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: change calling convention of existing cpumask APIs
Most cpumask functions started with cpus_: these have been replaced by
cpumask_ ones which take struct cpumask pointers as expected.
These four functions don't have good replacement names; fortunately
they're rarely used, so we just change them over.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: paulus@samba.org
Cc: mingo@redhat.com
Cc: tony.luck@intel.com
Cc: ralf@linux-mips.org
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Cc: cl@linux-foundation.org
Cc: srostedt@redhat.com
arch_reinit_sched_domains() used to call arch_update_cpu_topology()
via arch_init_sched_domains(). This call got lost with
e761b77252 ("cpu hotplug, sched: Introduce
cpu_active_map and redo sched domain managment (take 2)".
So we might end up with outdated and missing cpus in the cpu core
maps (architecture used to call arch_reinit_sched_domains if cpu
topology changed).
This adds a call to arch_update_cpu_topology in partition_sched_domains
which gets called whenever scheduling domains get updated. Which is
what is supposed to happen when cpu topology changes.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Change arch_update_cpu_topology so it returns 1 if the cpu topology changed
and 0 if it didn't change. This will be useful for the next patch which adds
a call to this function in partition_sched_domains.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The trace point only caught one of many places where a task changes cpu,
put it in the right place to we get all of them.
Change the signature while we're at it.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix possible deadlock in CPU hot-remove path
This patch fixes a possible deadlock scenario in the CPU remove path.
migration_call grabs rq->lock, then wakes up everything on rq->migration_queue
with the lock held. Then one of the tasks on the migration queue ends up
calling tg_shares_up which then also tries to acquire the same rq->lock.
[c000000058eab2e0] c000000000502078 ._spin_lock_irqsave+0x98/0xf0
[c000000058eab370] c00000000008011c .tg_shares_up+0x10c/0x20c
[c000000058eab430] c00000000007867c .walk_tg_tree+0xc4/0xfc
[c000000058eab4d0] c0000000000840c8 .try_to_wake_up+0xb0/0x3c4
[c000000058eab590] c0000000000799a0 .__wake_up_common+0x6c/0xe0
[c000000058eab640] c00000000007ada4 .complete+0x54/0x80
[c000000058eab6e0] c000000000509fa8 .migration_call+0x5fc/0x6f8
[c000000058eab7c0] c000000000504074 .notifier_call_chain+0x68/0xe0
[c000000058eab860] c000000000506568 ._cpu_down+0x2b0/0x3f4
[c000000058eaba60] c000000000506750 .cpu_down+0xa4/0x108
[c000000058eabb10] c000000000507e54 .store_online+0x44/0xa8
[c000000058eabba0] c000000000396260 .sysdev_store+0x3c/0x50
[c000000058eabc10] c0000000001a39b8 .sysfs_write_file+0x124/0x18c
[c000000058eabcd0] c00000000013061c .vfs_write+0xd0/0x1bc
[c000000058eabd70] c0000000001308a4 .sys_write+0x68/0x114
[c000000058eabe30] c0000000000086b4 syscall_exit+0x0/0x40
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix SD_BALANCE_NEWIDLEand broaden its use
load_balance_newidle() does not get called if SD_BALANCE_NEWIDLE is
set at higher level domain (3-CPU) and not in low level domain (2-MC).
pulled_task is initialised to -1 and checked for non-zero which is
always true if the lowest level sched_domain does not have
SD_BALANCE_NEWIDLE flag set.
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Implement the core kernel bits of Performance Counters subsystem.
The Linux Performance Counter subsystem provides an abstraction of
performance counter hardware capabilities. It provides per task and per
CPU counters, and it provides event capabilities on top of those.
Performance counters are accessed via special file descriptors.
There's one file descriptor per virtual counter used.
The special file descriptor is opened via the perf_counter_open()
system call:
int
perf_counter_open(u32 hw_event_type,
u32 hw_event_period,
u32 record_type,
pid_t pid,
int cpu);
The syscall returns the new fd. The fd can be used via the normal
VFS system calls: read() can be used to read the counter, fcntl()
can be used to set the blocking mode, etc.
Multiple counters can be kept open at a time, and the counters
can be poll()ed.
See more details in Documentation/perf-counters.txt.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: optimize the sched domains tree some more
The addition of SD_SERIALIZE flag added to SD_NODE_INIT prevented top level
dummy numa sched_domain to be properly degenerated on non-numa smp machine.
The reason is that in sd_parent_degenerate(), it found that the child and
parent does not have comon sched_domain flags due to SD_SERIALIZE. However,
for non-numa smp box, the top level is a dummy with a single sched_group.
Filter out SD_SERIALIZE if it is on non-numa machine to properly degenerate
top level node sched_domain. this will cut back some of the sd domain walk
in the load balancer code.
Signed-off-by: Ken Chen <kenchen@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Conflicts:
fs/nfsd/nfs4recover.c
Manually fixed above to use new creds API functions, e.g.
nfs4_save_creds().
Signed-off-by: James Morris <jmorris@namei.org>
Impact: extend information in /proc/sched_debug
This patch adds uid information in sched_debug for CONFIG_USER_SCHED
Signed-off-by: Arun R Bharadwaj <arun@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Regarding the bug addressed in:
4cd4262: sched: prevent divide by zero error in cpu_avg_load_per_task
Linus points out that the fix is not complete:
> There's nothing that keeps gcc from deciding not to reload
> rq->nr_running.
>
> Of course, in _practice_, I don't think gcc ever will (if it decides
> that it will spill, gcc is likely going to decide that it will
> literally spill the local variable to the stack rather than decide to
> reload off the pointer), but it's a valid compiler optimization, and
> it even has a name (rematerialization).
>
> So I suspect that your patch does fix the bug, but it still leaves the
> fairly unlikely _potential_ for it to re-appear at some point.
>
> We have ACCESS_ONCE() as a macro to guarantee that the compiler
> doesn't rematerialize a pointer access. That also would clarify
> the fact that we access something unsafe outside a lock.
So make sure our nr_running value is immutable and cannot change
after we check it for nonzero.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Move double_lock_balance()/double_unlock_balance() higher to fix the following
with gcc-3.4.6:
CC kernel/sched.o
In file included from kernel/sched.c:1605:
kernel/sched_rt.c: In function `find_lock_lowest_rq':
kernel/sched_rt.c:914: sorry, unimplemented: inlining failed in call to 'double_unlock_balance': function body not available
kernel/sched_rt.c:1077: sorry, unimplemented: called from here
make[2]: *** [kernel/sched.o] Error 1
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix divide by zero crash in scheduler rebalance irq
While testing the branch profiler, I hit this crash:
divide error: 0000 [#1] PREEMPT SMP
[...]
RIP: 0010:[<ffffffff8024a008>] [<ffffffff8024a008>] cpu_avg_load_per_task+0x50/0x7f
[...]
Call Trace:
<IRQ> <0> [<ffffffff8024fd43>] find_busiest_group+0x3e5/0xcaa
[<ffffffff8025da75>] rebalance_domains+0x2da/0xa21
[<ffffffff80478769>] ? find_next_bit+0x1b2/0x1e6
[<ffffffff8025e2ce>] run_rebalance_domains+0x112/0x19f
[<ffffffff8026d7c2>] __do_softirq+0xa8/0x232
[<ffffffff8020ea7c>] call_softirq+0x1c/0x3e
[<ffffffff8021047a>] do_softirq+0x94/0x1cd
[<ffffffff8026d5eb>] irq_exit+0x6b/0x10e
[<ffffffff8022e6ec>] smp_apic_timer_interrupt+0xd3/0xff
[<ffffffff8020e4b3>] apic_timer_interrupt+0x13/0x20
The code for cpu_avg_load_per_task has:
if (rq->nr_running)
rq->avg_load_per_task = rq->load.weight / rq->nr_running;
The runqueue lock is not held here, and there is nothing that prevents
the rq->nr_running from going to zero after it passes the if condition.
The branch profiler simply made the race window bigger.
This patch saves off the rq->nr_running to a local variable and uses that
for both the condition and the division.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: locking fix
We can't call cpuset_cpus_allowed_locked() with the rq lock held.
However, the rq lock merely protects us from (1) cpu_online_mask changing
and (2) someone else changing p->cpus_allowed.
The first can't happen because we're being called from a cpu hotplug
notifier. The second doesn't really matter: we are forcing the task off
a CPU it was affine to, so we're not doing very well anyway.
So we remove the rq lock from this path, and all is good.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
This patch changes the name of the "return function tracer" into
function-graph-tracer which is a more suitable name for a tracing
which makes one able to retrieve the ordered call stack during
the code flow.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup, move all hrtimer processing into hardirq context
This is an attempt at removing some of the hrtimer complexity by
reducing the number of callback modes to 1.
This means that all hrtimer callback functions will be ran from HARD-irq
context.
I went through all the 30 odd hrtimer callback functions in the kernel
and saw only one that I'm not quite sure of, which is the one in
net/can/bcm.c - hence I'm CC-ing the folks responsible for that code.
Furthermore, the hrtimer core now calls callbacks directly with IRQs
disabled in case you try to enqueue an expired timer. If this timer is a
periodic timer (which should use hrtimer_forward() to advance its time)
then it might be possible to end up in an inf. recursive loop due to the
fact that hrtimer_forward() doesn't round up to the next timer
granularity, and therefore keeps on calling the callback - obviously
this needs a fix.
Aside from that, this seems to compile and actually boot on my dual core
test box - although I'm sure there are some bugs in, me not hitting any
makes me certain :-)
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: Trivial API conversion
NR_CPUS -> nr_cpu_ids
cpumask_t -> struct cpumask
sizeof(cpumask_t) -> cpumask_size()
cpumask_a = cpumask_b -> cpumask_copy(&cpumask_a, &cpumask_b)
cpu_set() -> cpumask_set_cpu()
first_cpu() -> cpumask_first()
cpumask_of_cpu() -> cpumask_of()
cpus_* -> cpumask_*
There are some FIXMEs where we all archs to complete infrastructure
(patches have been sent):
cpu_coregroup_map -> cpu_coregroup_mask
node_to_cpumask* -> cpumask_of_node
There is also one FIXME where we pass an array of cpumasks to
partition_sched_domains(): this implies knowing the definition of
'struct cpumask' and the size of a cpumask. This will be fixed in a
future patch.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: (future) size reduction for large NR_CPUS.
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space for small nr_cpu_ids but big CONFIG_NR_CPUS. cpumask_var_t
is just a struct cpumask for !CONFIG_CPUMASK_OFFSTACK.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction, (future) size reduction for large NR_CPUS.
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space for small nr_cpu_ids but big CONFIG_NR_CPUS.
The fact cpupro_init is called both before and after the slab is
available makes for an ugly parameter unfortunately.
We also use cpumask_any_and to get rid of a temporary in cpupri_find.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: (future) size reduction for large NR_CPUS.
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space for small nr_cpu_ids but big CONFIG_NR_CPUS. cpumask_var_t
is just a struct cpumask for !CONFIG_CPUMASK_OFFSTACK.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction, (future) size reduction, cleanup
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space for small nr_cpu_ids but big CONFIG_NR_CPUS. cpumask_var_t
is just a struct cpumask for !CONFIG_CPUMASK_OFFSTACK.
We can also use cpulist_parse() instead of doing it manually in
isolated_cpu_setup.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
stack space. cpumask_var_t is just a struct cpumask for
!CONFIG_CPUMASK_OFFSTACK.
In this case, we always alloced, but we don't need to any more.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space on the stack. cpumask_var_t is just a struct cpumask for
!CONFIG_CPUMASK_OFFSTACK.
Note the removal of the initializer of new_mask: since the first thing
we did was "cpus_and(new_mask, new_mask, cpus_allowed)" I just changed
that to "cpumask_and(new_mask, in_mask, cpus_allowed);".
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction
With some care, we can avoid needing a temporary cpumask (we can't
really allocate here, since we can't fail).
This version calls cpuset_cpus_allowed_locked() with the task_rq_lock
held. I'm fairly sure this works, but there might be a deadlock
hiding.
And of course, we can't get rid of the last cpumask on stack until we
can use cpumask_of_node instead of node_to_cpumask.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space in the stack. cpumask_var_t is just a struct cpumask for
!CONFIG_CPUMASK_OFFSTACK.
Some jiggling here to make sure we always exit at the bottom (so we hit
the free_cpumask_var there).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space in the stack. cpumask_var_t is just a struct cpumask for
!CONFIG_CPUMASK_OFFSTACK.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: stack usage reduction
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space in the stack. cpumask_var_t is just a struct cpumask for
!CONFIG_CPUMASK_OFFSTACK.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: (future) size reduction for large NR_CPUS.
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space for small nr_cpu_ids but big CONFIG_NR_CPUS. cpumask_var_t
is just a struct cpumask for !CONFIG_CPUMASK_OFFSTACK.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: (future) size reduction for large NR_CPUS.
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space for small nr_cpu_ids but big CONFIG_NR_CPUS. cpumask_var_t
is just a struct cpumask for !CONFIG_CPUMASK_OFFSTACK.
def_root_domain is static, and so its masks are initialized with
alloc_bootmem_cpumask_var. After that, alloc_cpumask_var is used.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: (future) size reduction for large NR_CPUS.
Dynamically allocating cpumasks (when CONFIG_CPUMASK_OFFSTACK) saves
space for small nr_cpu_ids but big CONFIG_NR_CPUS. cpumask_var_t
is just a struct cpumask for !CONFIG_CPUMASK_OFFSTACK.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: (future) size reduction for large NR_CPUS.
We move the 'cpumask' member of sched_group to the end, so when we
kmalloc it we can do a minimal allocation: saves space for small
nr_cpu_ids but big CONFIG_NR_CPUS. Similar trick for 'span' in
sched_domain.
This isn't quite as good as converting to a cpumask_var_t, as some
sched_groups are actually static, but it's safer: we don't have to
figure out where to call alloc_cpumask_var/free_cpumask_var.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: trivial wrap of member accesses
This eases the transition in the next patch.
We also get rid of a temporary cpumask in find_idlest_cpu() thanks to
for_each_cpu_and, and sched_balance_self() due to getting weight before
setting sd to NULL.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: use new API
any_online_cpu() is a good name, but it takes a cpumask_t, not a
pointer.
There are several places where any_online_cpu() doesn't really want a
mask arg at all. Replace all callers with cpumask_any() and
cpumask_any_and().
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: use new general API
Using lots of allocs rather than one big alloc is less efficient, but
who cares for this setup function?
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: trivial API conversion
This is a simple conversion, but note that for_each_cpu() terminates
with i >= nr_cpu_ids, not i == NR_CPUS like for_each_cpu_mask() did.
I don't convert all of them: sd->span changes in a later patch, so
change those iterators there rather than here.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
* use node_to_cpumask_ptr in place of node_to_cpumask to reduce stack
requirements in sched.c
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: use deeper function tracing depth safely
Some tests showed that function return tracing needed a more deeper depth
of function calls. But it could be unsafe to store these return addresses
to the stack.
So these arrays will now be allocated dynamically into task_struct of current
only when the tracer is activated.
Typical scheme when tracer is activated:
- allocate a return stack for each task in global list.
- fork: allocate the return stack for the newly created task
- exit: free return stack of current
- idle init: same as fork
I chose a default depth of 50. I don't have overruns anymore.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
This commit:
commit f7b4cddcc5
Author: Oleg Nesterov <oleg@tv-sign.ru>
Date: Tue Oct 16 23:30:56 2007 -0700
do CPU_DEAD migrating under read_lock(tasklist) instead of write_lock_irq(ta
Currently move_task_off_dead_cpu() is called under
write_lock_irq(tasklist). This means it can't use task_lock() which is
needed to improve migrating to take task's ->cpuset into account.
Change the code to call move_task_off_dead_cpu() with irqs enabled, and
change migrate_live_tasks() to use read_lock(tasklist).
...forgot to update the comment in front of move_task_off_dead_cpu.
Reference: http://lkml.org/lkml/2008/6/23/135
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: make load-balancing more consistent
In the update_shares() path leading to tg_shares_up(), the calculation of
per-cpu cfs_rq shares is rather erratic even under moderate task wake up
rate. The problem is that the per-cpu tg->cfs_rq load weight used in the
sd_rq_weight aggregation and actual redistribution of the cfs_rq->shares
are collected at different time. Under moderate system load, we've seen
quite a bit of variation on the cfs_rq->shares and ultimately wildly
affects sched_entity's load weight.
This patch caches the result of initial per-cpu load weight when doing the
sum calculation, and then pass it down to update_group_shares_cpu() for
redistributing per-cpu cfs_rq shares. This allows consistent total cfs_rq
shares across all CPUs. It also simplifies the rounding and zero load
weight check.
Signed-off-by: Ken Chen <kenchen@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: properly rebuild sched-domains on kmalloc() failure
When cpuset failed to generate sched domains due to kmalloc()
failure, the scheduler should fallback to the single partition
'fallback_doms' and rebuild sched domains, but now it only
destroys but not rebuilds sched domains.
The regression was introduced by:
| commit dfb512ec48
| Author: Max Krasnyansky <maxk@qualcomm.com>
| Date: Fri Aug 29 13:11:41 2008 -0700
|
| sched: arch_reinit_sched_domains() must destroy domains to force rebuild
After the above commit, partition_sched_domains(0, NULL, NULL) will
only destroy sched domains and partition_sched_domains(1, NULL, NULL)
will create the default sched domain.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Max Krasnyansky <maxk@qualcomm.com>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: remove unnecessary accounting call
I don't actually understand account_steal_time() and I failed to find the
commit which added account_group_system_time(), but this looks bogus.
In any case rq->idle must be single-threaded, so it can't have ->totals.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: API *CHANGE*. Must update all tracepoint users.
Add DEFINE_TRACE() to tracepoints to let them declare the tracepoint
structure in a single spot for all the kernel. It helps reducing memory
consumption, especially when declaring a lot of tracepoints, e.g. for
kmalloc tracing.
*API CHANGE WARNING*: now, DECLARE_TRACE() must be used in headers for
tracepoint declarations rather than DEFINE_TRACE(). This is the sane way
to do it. The name previously used was misleading.
Updates scheduler instrumentation to follow this API change.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Conflicts:
security/keys/internal.h
security/keys/process_keys.c
security/keys/request_key.c
Fixed conflicts above by using the non 'tsk' versions.
Signed-off-by: James Morris <jmorris@namei.org>
Use RCU to access another task's creds and to release a task's own creds.
This means that it will be possible for the credentials of a task to be
replaced without another task (a) requiring a full lock to read them, and (b)
seeing deallocated memory.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>
Separate the task security context from task_struct. At this point, the
security data is temporarily embedded in the task_struct with two pointers
pointing to it.
Note that the Alpha arch is altered as it refers to (E)UID and (E)GID in
entry.S via asm-offsets.
With comment fixes Signed-off-by: Marc Dionne <marc.c.dionne@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>
Wrap access to task credentials so that they can be separated more easily from
the task_struct during the introduction of COW creds.
Change most current->(|e|s|fs)[ug]id to current_(|e|s|fs)[ug]id().
Change some task->e?[ug]id to task_e?[ug]id(). In some places it makes more
sense to use RCU directly rather than a convenient wrapper; these will be
addressed by later patches.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-audit@redhat.com
Cc: containers@lists.linux-foundation.org
Cc: linux-mm@kvack.org
Signed-off-by: James Morris <jmorris@namei.org>
Maciej Rutecki reported:
> I have this bug during suspend to disk:
>
> [ 188.592151] Enabling non-boot CPUs ...
> [ 188.592151] SMP alternatives: switching to SMP code
> [ 188.666058] BUG: using smp_processor_id() in preemptible
> [00000000]
> code: suspend_to_disk/2934
> [ 188.666064] caller is native_sched_clock+0x2b/0x80
Which, as noted by Linus, was caused by me, via:
7cbaef9c "sched: optimize sched_clock() a bit"
Move the rq locking a bit earlier in the initialization sequence,
that will make the sched_clock() call in init_idle() non-preemptible.
Reported-by: Maciej Rutecki <maciej.rutecki@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix load balancer load average calculation accuracy
cpu_avg_load_per_task() returns a stale value when nr_running is 0.
It returns an older stale (caculated when nr_running was non zero) value.
This patch returns and sets rq->avg_load_per_task to zero when nr_running
is 0.
Compile and boot tested on a x86_64 box.
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: improve CPU time accounting of tasks under the cpu accounting controller
Add hierarchical accounting to cpu accounting controller and include
cpuacct documentation.
Currently, while charging the task's cputime to its accounting group,
the accounting group hierarchy isn't updated. This patch charges the cputime
of a task to its accounting group and all its parent accounting groups.
Reported-by: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Paul Menage <menage@google.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix hang/crash on ia64 under high load
This is ugly, but the simplest patch by far.
Unlike other similar routines, account_group_exec_runtime() could be
called "implicitly" from within scheduler after exit_notify(). This
means we can race with the parent doing release_task(), we can't just
check ->signal != NULL.
Change __exit_signal() to do spin_unlock_wait(&task_rq(tsk)->lock)
before __cleanup_signal() to make sure ->signal can't be freed under
task_rq(tsk)->lock. Note that task_rq_unlock_wait() doesn't care
about the case when tsk changes cpu/rq under us, this should be OK.
Thanks to Ingo who nacked my previous buggy patch.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Reported-by: Doug Chapman <doug.chapman@hp.com>
Impact: clean up and fix debug info printout
While looking over the sched_debug code I noticed that we printed the rq
schedstats for every cfs_rq, ammend this.
Also change nr_spead_over into an int, and fix a little buglet in
min_vruntime printing.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
The #if/#endif is ugly. Change SCHED_CPUMASK_ALLOC and
SCHED_CPUMASK_FREE to static inline functions.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix rare memory leak in the sched-domains manual reconfiguration code
In the failure path, rd is not attached to a sched domain,
so it causes a leak.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We have a test case which measures the variation in the amount of time
needed to perform a fixed amount of work on the preempt_rt kernel. We
started seeing deterioration in it's performance recently. The test
should never take more than 10 microseconds, but we started 5-10%
failure rate.
Using elimination method, we traced the problem to commit
1b12bbc747 (lockdep: re-annotate
scheduler runqueues).
When LOCKDEP is disabled, this patch only adds an additional function
call to double_unlock_balance(). Hence I inlined double_unlock_balance()
and the problem went away. Here is a patch to make this change.
Signed-off-by: Sripathi Kodi <sripathik@in.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: improve/change/fix wakeup-buddy scheduling
Currently we only have a forward looking buddy, that is, we prefer to
schedule to the task we last woke up, under the presumption that its
going to consume the data we just produced, and therefore will have
cache hot benefits.
This allows co-waking producer/consumer task pairs to run ahead of the
pack for a little while, keeping their cache warm. Without this, we
would interleave all pairs, utterly trashing the cache.
This patch introduces a backward looking buddy, that is, suppose that
in the above scenario, the consumer preempts the producer before it
can go to sleep, we will therefore miss the wakeup from consumer to
producer (its already running, after all), breaking the cycle and
reverting to the cache-trashing interleaved schedule pattern.
The backward buddy will try to schedule back to the task that woke us
up in case the forward buddy is not available, under the assumption
that the last task will be the one with the most cache hot task around
barring current.
This will basically allow a task to continue after it got preempted.
In order to avoid starvation, we allow either buddy to get wakeup_gran
ahead of the pack.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup, add debug check
It's wrong to make dattr_new = NULL if doms_new == NULL, it introduces
memory leak if dattr_new != NULL. Fortunately dattr_new is always NULL
in this case. So remove the code and add a sanity check.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
The sysctl has been unregistered by partition_sched_domains().
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
Just use the newly introduced sd->name.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
So handling of sched_features read is simplified.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
Remove checking parent == NULL. It won't be NULLL, because we dynamically
create sub task_group only, and sub task_group always has its parent.
(root task_group is statically defined)
Also replace kmalloc_node(GFP_ZERO) with kzalloc_node().
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Since we moved wakeup preemption back to virtual time, it makes sense to move
the buddy stuff back as well. The purpose of the buddy scheduling is to allow
a quickly scheduling pair of tasks to run away from the group as far as a
regular busy task would be allowed under wakeup preemption.
This has the advantage that the pair can ping-pong for a while, enjoying
cache-hotness. Without buddy scheduling other tasks would interleave destroying
the cache.
Also, it saves a word in cfs_rq.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In one of the group load balancer patches:
commit 408ed066b1
Author: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Fri Jun 27 13:41:28 2008 +0200
Subject: sched: hierarchical load vs find_busiest_group
The following change:
- if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >=
+ if (max_load - this_load + 2*busiest_load_per_task >=
busiest_load_per_task * imbn) {
made the condition always true, because imbn is [1,2].
Therefore, remove the 2*, and give the it a fair chance.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'v28-range-hrtimers-for-linus-v2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (37 commits)
hrtimers: add missing docbook comments to struct hrtimer
hrtimers: simplify hrtimer_peek_ahead_timers()
hrtimers: fix docbook comments
DECLARE_PER_CPU needs linux/percpu.h
hrtimers: fix typo
rangetimers: fix the bug reported by Ingo for real
rangetimer: fix BUG_ON reported by Ingo
rangetimer: fix x86 build failure for the !HRTIMERS case
select: fix alpha OSF wrapper
select: fix alpha OSF wrapper
hrtimer: peek at the timer queue just before going idle
hrtimer: make the futex() system call use the per process slack value
hrtimer: make the nanosleep() syscall use the per process slack
hrtimer: fix signed/unsigned bug in slack estimator
hrtimer: show the timer ranges in /proc/timer_list
hrtimer: incorporate feedback from Peter Zijlstra
hrtimer: add a hrtimer_start_range() function
hrtimer: another build fix
hrtimer: fix build bug found by Ingo
hrtimer: make select() and poll() use the hrtimer range feature
...
* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: disable the hrtick for now
sched: revert back to per-rq vruntime
sched: fair scheduler should not resched rt tasks
sched: optimize group load balancer
sched: minor fast-path overhead reduction
sched: fix the wrong mask_len, cleanup
sched: kill unused scheduler decl.
sched: fix the wrong mask_len
sched: only update rq->clock while holding rq->lock
I noticed that tg_shares_up() unconditionally takes rq-locks for all cpus
in the sched_domain. This hurts.
We need the rq-locks whenever we change the weight of the per-cpu group sched
entities. To allevate this a little, only change the weight when the new
weight is at least shares_thresh away from the old value.
This avoids the rq-lock for the top level entries, since those will never
be re-weighted, and fuzzes the lower level entries a little to gain performance
in semi-stable situations.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Vatsa noticed rq->clock going funny and tracked it down to an update_rq_clock()
outside a rq->lock section.
This is a problem because things like double_rq_lock() update the rq->clock
value for both rqs. Therefore disabling interrupts isn't strong enough.
Reported-by: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Instrument the scheduler activity (sched_switch, migration, wakeups,
wait for a task, signal delivery) and process/thread
creation/destruction (fork, exit, kthread stop). Actually, kthread
creation is not instrumented in this patch because it is architecture
dependent. It allows to connect tracers such as ftrace which detects
scheduling latencies, good/bad scheduler decisions. Tools like LTTng can
export this scheduler information along with instrumentation of the rest
of the kernel activity to perform post-mortem analysis on the scheduler
activity.
About the performance impact of tracepoints (which is comparable to
markers), even without immediate values optimizations, tests done by
Hideo Aoki on ia64 show no regression. His test case was using hackbench
on a kernel where scheduler instrumentation (about 5 events in code
scheduler code) was added. See the "Tracepoints" patch header for
performance result detail.
Changelog :
- Change instrumentation location and parameter to match ftrace
instrumentation, previously done with kernel markers.
[ mingo@elte.hu: conflict resolutions ]
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Acked-by: 'Peter Zijlstra' <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
add /proc/sys/kernel/sched_domain/cpu0/domain0/name, to make
it easier to see which specific scheduler domain remained at
that entry.
Since we process the scheduler domain tree and
simplify it, it's not always immediately clear during debugging
which domain came from where.
depends on CONFIG_SCHED_DEBUG=y.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
css will be initialized by cgroup core.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch helped me out with a problem I recently had....
Basically, when the kernel lock is held, then preempt_count underflow does not
get detected until it is released which may be a long time (and arbitrarily,
eg at different points it may be rescheduled). If the bkl is released at
schedule, the resulting output is actually fairly cryptic...
With any other lock that elevates preempt_count, it is illegal to schedule
under it (which would get found pretty quickly). bkl allows scheduling with
preempt_count elevated, which makes underflows hard to debug.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: per CPU hrtimers can be migrated from a dead CPU
The hrtimer code has no knowledge about per CPU timers, but we need to
prevent the migration of such timers and warn when such a timer is
active at migration time.
Explicitely mark the timers as per CPU and use a more understandable
mode descriptor for the interrupts safe unlocked callback mode, which
is used by hrtimer_sleeper and the scheduler code.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
- fix UP lockup
- another set of UP/SMP cleanups and simplifications
Signed-off-by: Frank Mayhar <fmayhar@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
While playing around with it, I noticed we missed some sanity checks.
Also add some comments while we're there.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This is the second resubmission of the posix timer rework patch, posted
a few days ago.
This includes the changes from the previous resubmittion, which addressed
Oleg Nesterov's comments, removing the RCU stuff from the patch and
un-inlining the thread_group_cputime() function for SMP.
In addition, per Ingo Molnar it simplifies the UP code, consolidating much
of it with the SMP version and depending on lower-level SMP/UP handling to
take care of the differences.
It also cleans up some UP compile errors, moves the scheduler stats-related
macros into kernel/sched_stats.h, cleans up a merge error in
kernel/fork.c and has a few other minor fixes and cleanups as suggested
by Oleg and Ingo. Thanks for the review, guys.
Signed-off-by: Frank Mayhar <fmayhar@google.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
- Add some comments to try to make the ifdef puzzle a bit clearer
- Explicitly inline one of the three init_hrtick() implementations.
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
LD kernel/built-in.o
WARNING: kernel/built-in.o(.text+0x326): Section mismatch in reference
from the function init_hrtick() to the variable
.cpuinit.data:hotplug_hrtick_nb.8
The function init_hrtick() references
the variable __cpuinitdata hotplug_hrtick_nb.8.
This is often because init_hrtick lacks a __cpuinitdata
annotation or the annotation of hotplug_hrtick_nb.8 is wrong.
Signed-off-by: Md.Rakib H. Mullick <rakib.mullick@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Lin Ming reported a 10% OLTP regression against 2.6.27-rc4.
The difference seems to come from different preemption agressiveness,
which affects the cache footprint of the workload and its effective
cache trashing.
Aggresively preempt a task if its avg overlap is very small, this should
avoid the task going to sleep and find it still running when we schedule
back to it - saving a wakeup.
Reported-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Overview
This patch reworks the handling of POSIX CPU timers, including the
ITIMER_PROF, ITIMER_VIRT timers and rlimit handling. It was put together
with the help of Roland McGrath, the owner and original writer of this code.
The problem we ran into, and the reason for this rework, has to do with using
a profiling timer in a process with a large number of threads. It appears
that the performance of the old implementation of run_posix_cpu_timers() was
at least O(n*3) (where "n" is the number of threads in a process) or worse.
Everything is fine with an increasing number of threads until the time taken
for that routine to run becomes the same as or greater than the tick time, at
which point things degrade rather quickly.
This patch fixes bug 9906, "Weird hang with NPTL and SIGPROF."
Code Changes
This rework corrects the implementation of run_posix_cpu_timers() to make it
run in constant time for a particular machine. (Performance may vary between
one machine and another depending upon whether the kernel is built as single-
or multiprocessor and, in the latter case, depending upon the number of
running processors.) To do this, at each tick we now update fields in
signal_struct as well as task_struct. The run_posix_cpu_timers() function
uses those fields to make its decisions.
We define a new structure, "task_cputime," to contain user, system and
scheduler times and use these in appropriate places:
struct task_cputime {
cputime_t utime;
cputime_t stime;
unsigned long long sum_exec_runtime;
};
This is included in the structure "thread_group_cputime," which is a new
substructure of signal_struct and which varies for uniprocessor versus
multiprocessor kernels. For uniprocessor kernels, it uses "task_cputime" as
a simple substructure, while for multiprocessor kernels it is a pointer:
struct thread_group_cputime {
struct task_cputime totals;
};
struct thread_group_cputime {
struct task_cputime *totals;
};
We also add a new task_cputime substructure directly to signal_struct, to
cache the earliest expiration of process-wide timers, and task_cputime also
replaces the it_*_expires fields of task_struct (used for earliest expiration
of thread timers). The "thread_group_cputime" structure contains process-wide
timers that are updated via account_user_time() and friends. In the non-SMP
case the structure is a simple aggregator; unfortunately in the SMP case that
simplicity was not achievable due to cache-line contention between CPUs (in
one measured case performance was actually _worse_ on a 16-cpu system than
the same test on a 4-cpu system, due to this contention). For SMP, the
thread_group_cputime counters are maintained as a per-cpu structure allocated
using alloc_percpu(). The timer functions update only the timer field in
the structure corresponding to the running CPU, obtained using per_cpu_ptr().
We define a set of inline functions in sched.h that we use to maintain the
thread_group_cputime structure and hide the differences between UP and SMP
implementations from the rest of the kernel. The thread_group_cputime_init()
function initializes the thread_group_cputime structure for the given task.
The thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the
out-of-line function thread_group_cputime_alloc_smp() to allocate and fill
in the per-cpu structures and fields. The thread_group_cputime_free()
function, also a no-op for UP, in SMP frees the per-cpu structures. The
thread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls
thread_group_cputime_alloc() if the per-cpu structures haven't yet been
allocated. The thread_group_cputime() function fills the task_cputime
structure it is passed with the contents of the thread_group_cputime fields;
in UP it's that simple but in SMP it must also safely check that tsk->signal
is non-NULL (if it is it just uses the appropriate fields of task_struct) and,
if so, sums the per-cpu values for each online CPU. Finally, the three
functions account_group_user_time(), account_group_system_time() and
account_group_exec_runtime() are used by timer functions to update the
respective fields of the thread_group_cputime structure.
Non-SMP operation is trivial and will not be mentioned further.
The per-cpu structure is always allocated when a task creates its first new
thread, via a call to thread_group_cputime_clone_thread() from copy_signal().
It is freed at process exit via a call to thread_group_cputime_free() from
cleanup_signal().
All functions that formerly summed utime/stime/sum_sched_runtime values from
from all threads in the thread group now use thread_group_cputime() to
snapshot the values in the thread_group_cputime structure or the values in
the task structure itself if the per-cpu structure hasn't been allocated.
Finally, the code in kernel/posix-cpu-timers.c has changed quite a bit.
The run_posix_cpu_timers() function has been split into a fast path and a
slow path; the former safely checks whether there are any expired thread
timers and, if not, just returns, while the slow path does the heavy lifting.
With the dedicated thread group fields, timers are no longer "rebalanced" and
the process_timer_rebalance() function and related code has gone away. All
summing loops are gone and all code that used them now uses the
thread_group_cputime() inline. When process-wide timers are set, the new
task_cputime structure in signal_struct is used to cache the earliest
expiration; this is checked in the fast path.
Performance
The fix appears not to add significant overhead to existing operations. It
generally performs the same as the current code except in two cases, one in
which it performs slightly worse (Case 5 below) and one in which it performs
very significantly better (Case 2 below). Overall it's a wash except in those
two cases.
I've since done somewhat more involved testing on a dual-core Opteron system.
Case 1: With no itimer running, for a test with 100,000 threads, the fixed
kernel took 1428.5 seconds, 513 seconds more than the unfixed system,
all of which was spent in the system. There were twice as many
voluntary context switches with the fix as without it.
Case 2: With an itimer running at .01 second ticks and 4000 threads (the most
an unmodified kernel can handle), the fixed kernel ran the test in
eight percent of the time (5.8 seconds as opposed to 70 seconds) and
had better tick accuracy (.012 seconds per tick as opposed to .023
seconds per tick).
Case 3: A 4000-thread test with an initial timer tick of .01 second and an
interval of 10,000 seconds (i.e. a timer that ticks only once) had
very nearly the same performance in both cases: 6.3 seconds elapsed
for the fixed kernel versus 5.5 seconds for the unfixed kernel.
With fewer threads (eight in these tests), the Case 1 test ran in essentially
the same time on both the modified and unmodified kernels (5.2 seconds versus
5.8 seconds). The Case 2 test ran in about the same time as well, 5.9 seconds
versus 5.4 seconds but again with much better tick accuracy, .013 seconds per
tick versus .025 seconds per tick for the unmodified kernel.
Since the fix affected the rlimit code, I also tested soft and hard CPU limits.
Case 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer
running), the modified kernel was very slightly favored in that while
it killed the process in 19.997 seconds of CPU time (5.002 seconds of
wall time), only .003 seconds of that was system time, the rest was
user time. The unmodified kernel killed the process in 20.001 seconds
of CPU (5.014 seconds of wall time) of which .016 seconds was system
time. Really, though, the results were too close to call. The results
were essentially the same with no itimer running.
Case 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds
(where the hard limit would never be reached) and an itimer running,
the modified kernel exhibited worse tick accuracy than the unmodified
kernel: .050 seconds/tick versus .028 seconds/tick. Otherwise,
performance was almost indistinguishable. With no itimer running this
test exhibited virtually identical behavior and times in both cases.
In times past I did some limited performance testing. those results are below.
On a four-cpu Opteron system without this fix, a sixteen-thread test executed
in 3569.991 seconds, of which user was 3568.435s and system was 1.556s. On
the same system with the fix, user and elapsed time were about the same, but
system time dropped to 0.007 seconds. Performance with eight, four and one
thread were comparable. Interestingly, the timer ticks with the fix seemed
more accurate: The sixteen-thread test with the fix received 149543 ticks
for 0.024 seconds per tick, while the same test without the fix received 58720
for 0.061 seconds per tick. Both cases were configured for an interval of
0.01 seconds. Again, the other tests were comparable. Each thread in this
test computed the primes up to 25,000,000.
I also did a test with a large number of threads, 100,000 threads, which is
impossible without the fix. In this case each thread computed the primes only
up to 10,000 (to make the runtime manageable). System time dominated, at
1546.968 seconds out of a total 2176.906 seconds (giving a user time of
629.938s). It received 147651 ticks for 0.015 seconds per tick, still quite
accurate. There is obviously no comparable test without the fix.
Signed-off-by: Frank Mayhar <fmayhar@google.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
What I realized recently is that calling rebuild_sched_domains() in
arch_reinit_sched_domains() by itself is not enough when cpusets are enabled.
partition_sched_domains() code is trying to avoid unnecessary domain rebuilds
and will not actually rebuild anything if new domain masks match the old ones.
What this means is that doing
echo 1 > /sys/devices/system/cpu/sched_mc_power_savings
on a system with cpusets enabled will not take affect untill something changes
in the cpuset setup (ie new sets created or deleted).
This patch fixes restore correct behaviour where domains must be rebuilt in
order to enable MC powersaving flags.
Test on quad-core Core2 box with both CONFIG_CPUSETS and !CONFIG_CPUSETS.
Also tested on dual-core Core2 laptop. Lockdep is happy and things are working
as expected.
Signed-off-by: Max Krasnyansky <maxk@qualcomm.com>
Tested-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
I found that 2.6.27-rc5-mm1 does not compile with gcc 3.4.6.
The error is:
CC kernel/sched.o
kernel/sched.c: In function `start_rt_bandwidth':
kernel/sched.c:208: sorry, unimplemented: inlining failed in call to 'rt_bandwidth_enabled': function body not available
kernel/sched.c:214: sorry, unimplemented: called from here
make[1]: *** [kernel/sched.o] Error 1
make: *** [kernel] Error 2
It seems that the gcc 3.4.6 requires full inline definition before first usage.
The patch below fixes the compilation problem.
Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl> (if needed>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to be able to do range hrtimers we need to use accessor functions
to the "expire" member of the hrtimer struct.
This patch converts kernel/* to these accessors.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Spencer reported a problem where utime and stime were going negative despite
the fixes in commit b27f03d4bd. The suspected
reason for the problem is that signal_struct maintains it's own utime and
stime (of exited tasks), these are not updated using the new task_utime()
routine, hence sig->utime can go backwards and cause the same problem
to occur (sig->utime, adds tsk->utime and not task_utime()). This patch
fixes the problem
TODO: using max(task->prev_utime, derived utime) works for now, but a more
generic solution is to implement cputime_max() and use the cputime_gt()
function for comparison.
Reported-by: spencer@bluehost.com
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
fix:
kernel/sched.c: In function '__rt_schedulable':
kernel/sched.c:8771: error: implicit declaration of function 'walk_tg_tree'
kernel/sched.c:8771: error: 'tg_nop' undeclared (first use in this function)
kernel/sched.c:8771: error: (Each undeclared identifier is reported only once
kernel/sched.c:8771: error: for each function it appears in.)
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch adds kernel doc for the completion feature.
An error in the split-man.pl PERL snippet in kernel-doc-nano-HOWTO.txt is
also fixed.
Signed-off-by: Kevin Diggs <kevdig@hypersurf.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
wait_task_inactive() returns 1 when p->nvcsw == 0 || p->nvcsw == 1. This
means that two subsequent calls can return the same number while the task
was scheduled in between.
Change the code to return "nvcsw | LONG_MIN" instead of "nvcsw ?: 1", now
the overlap always needs LONG_MAX schedules.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
If wait_task_inactive() returns success the task was deactivated. In that
case schedule() always increments ->nvcsw which alone can be used as a
"generation counter".
If the next call returns the same number, we can be sure that the task was
unscheduled. Otherwise, because we know that .on_rq == 0 again, ->nvcsw
should have been changed in between.
Q: perhaps it is better to do "ncsw = (p->nvcsw << 1) | 1" ? This
decreases the possibility of "was it unscheduled" false positive when
->nvcsw == 0.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Change do_wait_for_common() to use signal_pending_state() instead of open
coding.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
