s390/powerpc/ia64 have support for CONFIG_VIRT_CPU_ACCOUNTING which does
the fine granularity accounting of user, system, hardirq, softirq times.
Adding that option on archs like x86 will be challenging however, given the
state of TSC reliability on various platforms and also the overhead it will
add in syscall entry exit.
Instead, add a lighter variant that only does finer accounting of
hardirq and softirq times, providing precise irq times (instead of timer tick
based samples). This accounting is added with a new config option
CONFIG_IRQ_TIME_ACCOUNTING so that there won't be any overhead for users not
interested in paying the perf penalty.
This accounting is based on sched_clock, with the code being generic.
So, other archs may find it useful as well.
This patch just adds the core logic and does not enable this logic yet.
Signed-off-by: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1286237003-12406-5-git-send-email-venki@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
To account softirq time cleanly in scheduler, we need to identify whether
softirq is invoked in ksoftirqd context or softirq at hardirq tail context.
Add PF_KSOFTIRQD for that purpose.
As all PF flag bits are currently taken, create space by moving one of the
infrequently used bits (PF_THREAD_BOUND) down in task_struct to be along
with some other state fields.
Signed-off-by: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1286237003-12406-4-git-send-email-venki@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra found a bug in the way softirq time is accounted in
VIRT_CPU_ACCOUNTING on this thread:
http://lkml.indiana.edu/hypermail//linux/kernel/1009.2/01366.html
The problem is, softirq processing uses local_bh_disable internally. There
is no way, later in the flow, to differentiate between whether softirq is
being processed or is it just that bh has been disabled. So, a hardirq when bh
is disabled results in time being wrongly accounted as softirq.
Looking at the code a bit more, the problem exists in !VIRT_CPU_ACCOUNTING
as well. As account_system_time() in normal tick based accouting also uses
softirq_count, which will be set even when not in softirq with bh disabled.
Peter also suggested solution of using 2*SOFTIRQ_OFFSET as irq count
for local_bh_{disable,enable} and using just SOFTIRQ_OFFSET while softirq
processing. The patch below does that and adds API in_serving_softirq() which
returns whether we are currently processing softirq or not.
Also changes one of the usages of softirq_count in net/sched/cls_cgroup.c
to in_serving_softirq.
Looks like many usages of in_softirq really want in_serving_softirq. Those
changes can be made individually on a case by case basis.
Signed-off-by: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1286237003-12406-2-git-send-email-venki@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When SD_PREFER_SIBLING is set on a sched domain, drop group_capacity to 1
only if the local group has extra capacity. The extra check prevents the case
where you always pull from the heaviest group when it is already under-utilized
(possible with a large weight task outweighs the tasks on the system).
For example, consider a 16-cpu quad-core quad-socket machine with MC and NUMA
scheduling domains. Let's say we spawn 15 nice0 tasks and one nice-15 task,
and each task is running on one core. In this case, we observe the following
events when balancing at the NUMA domain:
- find_busiest_group() will always pick the sched group containing the niced
task to be the busiest group.
- find_busiest_queue() will then always pick one of the cpus running the
nice0 task (never picks the cpu with the nice -15 task since
weighted_cpuload > imbalance).
- The load balancer fails to migrate the task since it is the running task
and increments sd->nr_balance_failed.
- It repeats the above steps a few more times until sd->nr_balance_failed > 5,
at which point it kicks off the active load balancer, wakes up the migration
thread and kicks the nice 0 task off the cpu.
The load balancer doesn't stop until we kick out all nice 0 tasks from
the sched group, leaving you with 3 idle cpus and one cpu running the
nice -15 task.
When balancing at the NUMA domain, we drop sgs.group_capacity to 1 if the child
domain (in this case MC) has SD_PREFER_SIBLING set. Subsequent load checks are
not relevant because the niced task has a very large weight.
In this patch, we add an extra condition to the "if(prefer_sibling)" check in
update_sd_lb_stats(). We drop the capacity of a group only if the local group
has extra capacity, ie. nr_running < group_capacity. This patch preserves the
original intent of the prefer_siblings check (to spread tasks across the system
in low utilization scenarios) and fixes the case above.
It helps in the following ways:
- In low utilization cases (where nr_tasks << nr_cpus), we still drop
group_capacity down to 1 if we prefer siblings.
- On very busy systems (where nr_tasks >> nr_cpus), sgs.nr_running will most
likely be > sgs.group_capacity.
- When balancing large weight tasks, if the local group does not have extra
capacity, we do not pick the group with the niced task as the busiest group.
This prevents failed balances, active migration and the under-utilization
described above.
Signed-off-by: Nikhil Rao <ncrao@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1287173550-30365-5-git-send-email-ncrao@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch forces a load balance on a newly idle cpu when the local group has
extra capacity and the busiest group does not have any. It improves system
utilization when balancing tasks with a large weight differential.
Under certain situations, such as a niced down task (i.e. nice = -15) in the
presence of nr_cpus NICE0 tasks, the niced task lands on a sched group and
kicks away other tasks because of its large weight. This leads to sub-optimal
utilization of the machine. Even though the sched group has capacity, it does
not pull tasks because sds.this_load >> sds.max_load, and f_b_g() returns NULL.
With this patch, if the local group has extra capacity, we shortcut the checks
in f_b_g() and try to pull a task over. A sched group has extra capacity if the
group capacity is greater than the number of running tasks in that group.
Thanks to Mike Galbraith for discussions leading to this patch and for the
insight to reuse SD_NEWIDLE_BALANCE.
Signed-off-by: Nikhil Rao <ncrao@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1287173550-30365-4-git-send-email-ncrao@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When cycling through sched groups to determine the busiest group, set
group_imb only if the busiest cpu has more than 1 runnable task. This patch
fixes the case where two cpus in a group have one runnable task each, but there
is a large weight differential between these two tasks. The load balancer is
unable to migrate any task from this group, and hence do not consider this
group to be imbalanced.
Signed-off-by: Nikhil Rao <ncrao@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1286996978-7007-3-git-send-email-ncrao@google.com>
[ small code readability edits ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch adds a check in task_hot to return if the task has SCHED_IDLE
policy. SCHED_IDLE tasks have very low weight, and when run with regular
workloads, are typically scheduled many milliseconds apart. There is no
need to consider these tasks hot for load balancing.
Signed-off-by: Nikhil Rao <ncrao@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1287173550-30365-2-git-send-email-ncrao@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Load weights are for the CFS, they do not belong in the RT task. This makes all
RT scheduling classes leave the CFS weights alone.
This fixes a real bug as well: I noticed the following phonomena: a process
elevated to SCHED_RR forks with SCHED_RESET_ON_FORK set, and the child is
indeed SCHED_OTHER, and the niceval is indeed reset to 0. However the weight
inserted by set_load_weight() remains at 0, giving the task insignificat
priority.
With this fix, the weight is reset to what the task had before being elevated
to SCHED_RR/SCHED_FIFO.
Cc: Lennart Poettering <lennart@poettering.net>
Cc: stable@kernel.org
Signed-off-by: Linus Walleij <linus.walleij@stericsson.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1286807811-10568-1-git-send-email-linus.walleij@stericsson.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to separate the stop/migrate work thread from the SCHED_FIFO
implementation, create a special class for it that is of higher priority than
SCHED_FIFO itself.
This currently solves a problem where cpu-hotplug consumes so much cpu-time
that the SCHED_FIFO class gets throttled, but has the bandwidth replenishment
timer pending on the now dead cpu.
It is also required for when we add the planned deadline scheduling class above
SCHED_FIFO, as the stop/migrate thread still needs to transcent those tasks.
Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1285165776.2275.1022.camel@laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Labels should be on column 0.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Time stamps for the ring buffer are created by the difference between
two events. Each page of the ring buffer holds a full 64 bit timestamp.
Each event has a 27 bit delta stamp from the last event. The unit of time
is nanoseconds, so 27 bits can hold ~134 milliseconds. If two events
happen more than 134 milliseconds apart, a time extend is inserted
to add more bits for the delta. The time extend has 59 bits, which
is good for ~18 years.
Currently the time extend is committed separately from the event.
If an event is discarded before it is committed, due to filtering,
the time extend still exists. If all events are being filtered, then
after ~134 milliseconds a new time extend will be added to the buffer.
This can only happen till the end of the page. Since each page holds
a full timestamp, there is no reason to add a time extend to the
beginning of a page. Time extends can only fill a page that has actual
data at the beginning, so there is no fear that time extends will fill
more than a page without any data.
When reading an event, a loop is made to skip over time extends
since they are only used to maintain the time stamp and are never
given to the caller. As a paranoid check to prevent the loop running
forever, with the knowledge that time extends may only fill a page,
a check is made that tests the iteration of the loop, and if the
iteration is more than the number of time extends that can fit in a page
a warning is printed and the ring buffer is disabled (all of ftrace
is also disabled with it).
There is another event type that is called a TIMESTAMP which can
hold 64 bits of data in the theoretical case that two events happen
18 years apart. This code has not been implemented, but the name
of this event exists, as well as the structure for it. The
size of a TIMESTAMP is 16 bytes, where as a time extend is only
8 bytes. The macro used to calculate how many time extends can fit on
a page used the TIMESTAMP size instead of the time extend size
cutting the amount in half.
The following test case can easily trigger the warning since we only
need to have half the page filled with time extends to trigger the
warning:
# cd /sys/kernel/debug/tracing/
# echo function > current_tracer
# echo 'common_pid < 0' > events/ftrace/function/filter
# echo > trace
# echo 1 > trace_marker
# sleep 120
# cat trace
Enabling the function tracer and then setting the filter to only trace
functions where the process id is negative (no events), then clearing
the trace buffer to ensure that we have nothing in the buffer,
then write to trace_marker to add an event to the beginning of a page,
sleep for 2 minutes (only 35 seconds is probably needed, but this
guarantees the bug), and then finally reading the trace which will
trigger the bug.
This patch fixes the typo and prevents the false positive of that warning.
Reported-by: Hans J. Koch <hjk@linutronix.de>
Tested-by: Hans J. Koch <hjk@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Stable Kernel <stable@kernel.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
* 'hwpoison-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-mce-2.6:
HWPOISON: Stop shrinking at right page count
HWPOISON: Report correct address granuality for AO huge page errors
HWPOISON: Copy si_addr_lsb to user
page-types.c: fix name of unpoison interface
When proc_doulongvec_minmax() is used with an array of longs, and no
min/max check requested (.extra1 or .extra2 being NULL), we dereference a
NULL pointer for the second element of the array.
Noticed while doing some changes in network stack for the "16TB problem"
Fix is to not change min & max pointers in __do_proc_doulongvec_minmax(),
so that all elements of the vector share an unique min/max limit, like
proc_dointvec_minmax().
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Americo Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The original hwpoison code added a new siginfo field si_addr_lsb to
pass the granuality of the fault address to user space. Unfortunately
this field was never copied to user space. Fix this here.
I added explicit checks for the MCEERR codes to avoid having
to patch all potential callers to initialize the field.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
With all the recent module loading cleanups, we've minimized the code
that sits under module_mutex, fixing various deadlocks and making it
possible to do most of the module loading in parallel.
However, that whole conversion totally missed the rather obscure code
that adds a new module to the list for BUG() handling. That code was
doubly obscure because (a) the code itself lives in lib/bugs.c (for
dubious reasons) and (b) it gets called from the architecture-specific
"module_finalize()" rather than from generic code.
Calling it from arch-specific code makes no sense what-so-ever to begin
with, and is now actively wrong since that code isn't protected by the
module loading lock any more.
So this commit moves the "module_bug_{finalize,cleanup}()" calls away
from the arch-specific code, and into the generic code - and in the
process protects it with the module_mutex so that the list operations
are now safe.
Future fixups:
- move the module list handling code into kernel/module.c where it
belongs.
- get rid of 'module_bug_list' and just use the regular list of modules
(called 'modules' - imagine that) that we already create and maintain
for other reasons.
Reported-and-tested-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Adrian Bunk <bunk@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The kfifo_dma family of functions use sg_mark_end() on the last element in
their scatterlist. This forces use of a fresh scatterlist for each DMA
operation, which makes recycling a single scatterlist impossible.
Change the behavior of the kfifo_dma functions to match the usage of the
dma_map_sg function. This means that users must respect the returned
nents value. The sample code is updated to reflect the change.
This bug is trivial to cause: call kfifo_dma_in_prepare() such that it
prepares a scatterlist with a single entry comprising the whole fifo.
This is the case when you map the entirety of a newly created empty fifo.
This causes the setup_sgl() function to mark the first scatterlist entry
as the end of the chain, no matter what comes after it.
Afterwards, add and remove some data from the fifo such that another call
to kfifo_dma_in_prepare() will create two scatterlist entries. It returns
nents=2. However, due to the previous sg_mark_end() call, sg_is_last()
will now return true for the first scatterlist element. This causes the
sample code to print a single scatterlist element when it should print
two.
By removing the call to sg_mark_end(), we make the API as similar as
possible to the DMA mapping API. All users are required to respect the
returned nents.
Signed-off-by: Ira W. Snyder <iws@ovro.caltech.edu>
Cc: Stefani Seibold <stefani@seibold.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The below bug in fork led to the rmap walk finding the parent huge-pmd
twice instead of just once, because the anon_vma_chain objects of the
child vma still point to the vma->vm_mm of the parent.
The patch fixes it by making the rmap walk accurate during fork. It's not
a big deal normally but it worth being accurate considering the cost is
the same.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: Fix nohz balance kick
sched: Fix user time incorrectly accounted as system time on 32-bit
If a high priority task is waking up on a CPU that is running a
lower priority task that is bound to a CPU, see if we can move the
high RT task to another CPU first. Note, if all other CPUs are
running higher priority tasks than the CPU bounded current task,
then it will be preempted regardless.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Gregory Haskins <ghaskins@novell.com>
LKML-Reference: <20100921024138.888922071@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
