Currently even if BLKTRACESETUP ioctl has failed user must call
BLKTRACETEARDOWN to be shure what all staff was cleaned, which
is contr-intuitive.
Let's setup ioctl make necessery cleanup by it self.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
The function graph tracer is currently the most invasive tracer
in the ftrace family. It can easily overflow the buffer even with
10megs per CPU. This means that events can often be lost.
On start up, or after events are lost, if the function return is
recorded but the function enter was lost, all we get to see is the
exiting '}'.
Here is how a typical trace output starts:
[tracing] cat trace
# tracer: function_graph
#
# CPU DURATION FUNCTION CALLS
# | | | | | | |
0) + 91.897 us | }
0) ! 567.961 us | }
0) <========== |
0) ! 579.083 us | _raw_spin_lock_irqsave();
0) 4.694 us | _raw_spin_unlock_irqrestore();
0) ! 594.862 us | }
0) ! 603.361 us | }
0) ! 613.574 us | }
0) ! 623.554 us | }
0) 3.653 us | fget_light();
0) | sock_poll() {
There are a series of '}' with no matching "func() {". There's no information
to what functions these ending brackets belong to.
This patch adds a stack on the per cpu structure used in outputting
the function graph tracer to keep track of what function was outputted.
Then on a function exit event, it checks the depth to see if the
function exit has a matching entry event. If it does, then it only
prints the '}', otherwise it adds the function name after the '}'.
This allows function exit events to show what function they belong to
at trace output startup, when the entry was lost due to ring buffer
overflow, or even after a new task is scheduled in.
Here is what the above trace will look like after this patch:
[tracing] cat trace
# tracer: function_graph
#
# CPU DURATION FUNCTION CALLS
# | | | | | | |
0) + 91.897 us | } (irq_exit)
0) ! 567.961 us | } (smp_apic_timer_interrupt)
0) <========== |
0) ! 579.083 us | _raw_spin_lock_irqsave();
0) 4.694 us | _raw_spin_unlock_irqrestore();
0) ! 594.862 us | } (add_wait_queue)
0) ! 603.361 us | } (__pollwait)
0) ! 613.574 us | } (tcp_poll)
0) ! 623.554 us | } (sock_poll)
0) 3.653 us | fget_light();
0) | sock_poll() {
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The hibernate memory preallocation code allocates memory to push some
user space data out of physical RAM, so that the hibernation image is
not too large. It allocates more memory than necessary for creating
the image, so it has to release some pages to make room for
allocations made while suspending devices and disabling nonboot CPUs,
or the system will hang due to the lack of free pages to allocate
from. Unfortunately, the function used for freeing these pages,
free_unnecessary_pages(), contains a bug that prevents it from doing
the job on all systems without highmem.
Fix this problem, which is a regression from the 2.6.30 kernel, by
using the right condition for the termination of the loop in
free_unnecessary_pages().
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Reported-and-tested-by: Alan Jenkins <sourcejedi.lkml@googlemail.com>
Cc: stable@kernel.org
Its contents and entry in Makefile were already removed in
8e60c6a134
(Shift remaining code from swsusp.c to hibernate.c)
but somehow it remained in-place (rjw: which most likely was my
mistake).
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Nigel Cunningham <nigel@tuxonice.net>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Remove a trailing space from a message in swsusp_save().
Signed-off-by: Frans Pop <elendil@planet.nl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
It will never reach here if the sws_resume_bdev is erratic.
swsusp_read() is called only from software_resume(), but after
swsusp_check() which would catch the error state.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
They were deprecated and removed from exported headers more than 2
years ago. Inform users about their removal in the future now.
(Switch cases needed to be reorderded for an easy fall through.)
And add an entry to feature-removal-schedule.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Add configuration switch CONFIG_PM_ADVANCED_DEBUG for compiling in
extra PM debugging/testing code allowing one to access some
PM-related attributes of devices from the user space via sysfs.
If CONFIG_PM_ADVANCED_DEBUG is set, add sysfs attribute power/async
for every device allowing the user space to access the device's
power.async_suspend flag and modify it, if desired.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Add sysfs attribute /sys/power/pm_async allowing the user space to
disable/enable asynchronous suspend/resume of devices.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
A recent commit introduced a preemption warning for
perf_clock(), use raw_smp_processor_id() to avoid this, it
really doesn't matter which cpu we use here.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1267198583.22519.684.camel@laptop>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
On platforms like dual socket quad-core platform, the scheduler load
balancer is not detecting the load imbalances in certain scenarios. This
is leading to scenarios like where one socket is completely busy (with
all the 4 cores running with 4 tasks) and leaving another socket
completely idle. This causes performance issues as those 4 tasks share
the memory controller, last-level cache bandwidth etc. Also we won't be
taking advantage of turbo-mode as much as we would like, etc.
Some of the comparisons in the scheduler load balancing code are
comparing the "weighted cpu load that is scaled wrt sched_group's
cpu_power" with the "weighted average load per task that is not scaled
wrt sched_group's cpu_power". While this has probably been broken for a
longer time (for multi socket numa nodes etc), the problem got aggrevated
via this recent change:
|
| commit f93e65c186
| Author: Peter Zijlstra <a.p.zijlstra@chello.nl>
| Date: Tue Sep 1 10:34:32 2009 +0200
|
| sched: Restore __cpu_power to a straight sum of power
|
Also with this change, the sched group cpu power alone no longer reflects
the group capacity that is needed to implement MC, MT performance
(default) and power-savings (user-selectable) policies.
We need to use the computed group capacity (sgs.group_capacity, that is
computed using the SD_PREFER_SIBLING logic in update_sd_lb_stats()) to
find out if the group with the max load is above its capacity and how
much load to move etc.
Reported-by: Ma Ling <ling.ma@intel.com>
Initial-Analysis-by: Zhang, Yanmin <yanmin_zhang@linux.intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
[ -v2: build fix ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <stable@kernel.org> # [2.6.32.x, 2.6.33.x]
LKML-Reference: <1266970432.11588.22.camel@sbs-t61.sc.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Since the cpu argument to hw_perf_group_sched_in() is always
smp_processor_id(), simplify the code a little by removing this argument
and using the current cpu where needed.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: David Miller <davem@davemloft.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1265890918.5396.3.camel@laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch adds correct AMD NorthBridge event scheduling.
NB events are events measuring L3 cache, Hypertransport traffic. They are
identified by an event code >= 0xe0. They measure events on the
Northbride which is shared by all cores on a package. NB events are
counted on a shared set of counters. When a NB event is programmed in a
counter, the data actually comes from a shared counter. Thus, access to
those counters needs to be synchronized.
We implement the synchronization such that no two cores can be measuring
NB events using the same counters. Thus, we maintain a per-NB allocation
table. The available slot is propagated using the event_constraint
structure.
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <4b703957.0702d00a.6bf2.7b7d@mx.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In certain situations, the kernel may need to stop and start the same
event rapidly. The current PMU callbacks do not distinguish between stop
and release (i.e., stop + free the resource). Thus, a counter may be
released, then it will be immediately re-acquired. Event scheduling will
again take place with no guarantee to assign the same counter. On some
processors, this may event yield to failure to assign the event back due
to competion between cores.
This patch is adding a new pair of callback to stop and restart a counter
without actually release the underlying counter resource. On stop, the
counter is stopped, its values saved and that's it. On start, the value
is reloaded and counter is restarted (on x86, actual restart is delayed
until perf_enable()).
Signed-off-by: Stephane Eranian <eranian@google.com>
[ added fallback to ->enable/->disable for all other PMUs
fixed x86_pmu_start() to call x86_pmu.enable()
merged __x86_pmu_disable into x86_pmu_stop() ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <4b703875.0a04d00a.7896.ffffb824@mx.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
DaveM reported that currently perf interprets the pgoff value reported by
the MMAP events as a byte range, but the kernel reports it as a page
offset.
Since its broken (and unusable) anyway, change the kernel behaviour (ABI)
to report bytes indeed, avoiding the need for userspace to deal with
PAGE_SIZE things.
Reported-by: David Miller <davem@davemloft.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Discard freeing field->type since it is not necessary.
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Wenji Huang <wenji.huang@oracle.com>
LKML-Reference: <1266997226-6833-5-git-send-email-wenji.huang@oracle.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The "cpu" variable is declared at the start of the function and
also within a branch, with the exact same initialization.
Remove the local variable of the same name in the branch.
Signed-off-by: Wenji Huang <wenji.huang@oracle.com>
LKML-Reference: <1266997226-6833-3-git-send-email-wenji.huang@oracle.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The power tracer has been converted to power trace events.
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <4B84D50E.4070806@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
GCC 4.5 introduces behavior that forces the alignment of structures to
use the largest possible value. The default value is 32 bytes, so if
some structures are defined with a 4-byte alignment and others aren't
declared with an alignment constraint at all - it will align at 32-bytes.
For things like the ftrace events, this results in a non-standard array.
When initializing the ftrace subsystem, we traverse the _ftrace_events
section and call the initialization callback for each event. When the
structures are misaligned, we could be treating another part of the
structure (or the zeroed out space between them) as a function pointer.
This patch forces the alignment for all the ftrace_event_call structures
to 4 bytes.
Without this patch, the kernel fails to boot very early when built with
gcc 4.5.
It's trivial to check the alignment of the members of the array, so it
might be worthwhile to add something to the build system to do that
automatically. Unfortunately, that only covers this case. I've asked one
of the gcc developers about adding a warning when this condition is seen.
Cc: stable@kernel.org
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
LKML-Reference: <4B85770B.6010901@suse.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Under TREE_PREEMPT_RCU, print_other_cpu_stall() invokes
rcu_print_task_stall() with the root rcu_node structure's ->lock
held, and rcu_print_task_stall() acquires that same lock for
self-deadlock. Fix this by removing the lock acquisition from
rcu_print_task_stall(), and making all callers acquire the lock
instead.
Tested-by: John Kacur <jkacur@redhat.com>
Tested-by: Thomas Gleixner <tglx@linutronix.de>
Located-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: laijs@cn.fujitsu.com
Cc: dipankar@in.ibm.com
Cc: mathieu.desnoyers@polymtl.ca
Cc: josh@joshtriplett.org
Cc: dvhltc@us.ibm.com
Cc: niv@us.ibm.com
Cc: peterz@infradead.org
Cc: rostedt@goodmis.org
Cc: Valdis.Kletnieks@vt.edu
Cc: dhowells@redhat.com
LKML-Reference: <1266887105-1528-19-git-send-email-paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The C standard does not specify the result of an operation that
overflows a signed integer, so such operations need to be
avoided. This patch changes the type of several fields from
"long" to "unsigned long" and adjusts operations as needed.
ULONG_CMP_GE() and ULONG_CMP_LT() macros are introduced to do
the modular comparisons that are appropriate given that overflow
is an expected event.
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: laijs@cn.fujitsu.com
Cc: dipankar@in.ibm.com
Cc: mathieu.desnoyers@polymtl.ca
Cc: josh@joshtriplett.org
Cc: dvhltc@us.ibm.com
Cc: niv@us.ibm.com
Cc: peterz@infradead.org
Cc: rostedt@goodmis.org
Cc: Valdis.Kletnieks@vt.edu
Cc: dhowells@redhat.com
LKML-Reference: <1266887105-1528-17-git-send-email-paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Currently, rcu_needs_cpu() simply checks whether the current CPU
has an outstanding RCU callback, which means that the last CPU
to go into dyntick-idle mode might wait a few ticks for the
relevant grace periods to complete. However, if all the other
CPUs are in dyntick-idle mode, and if this CPU is in a quiescent
state (which it is for RCU-bh and RCU-sched any time that we are
considering going into dyntick-idle mode), then the grace period
is instantly complete.
This patch therefore repeatedly invokes the RCU grace-period
machinery in order to force any needed grace periods to complete
quickly. It does so a limited number of times in order to
prevent starvation by an RCU callback function that might pass
itself to call_rcu().
However, if any CPU other than the current one is not in
dyntick-idle mode, fall back to simply checking (with fix to bug
noted by Lai Jiangshan). Also, take advantage of last
grace-period forcing, the opportunity to do so noted by Steve
Rostedt. And apply simplified #ifdef condition suggested by
Frederic Weisbecker.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: laijs@cn.fujitsu.com
Cc: dipankar@in.ibm.com
Cc: mathieu.desnoyers@polymtl.ca
Cc: josh@joshtriplett.org
Cc: dvhltc@us.ibm.com
Cc: niv@us.ibm.com
Cc: peterz@infradead.org
Cc: rostedt@goodmis.org
Cc: Valdis.Kletnieks@vt.edu
Cc: dhowells@redhat.com
LKML-Reference: <1266887105-1528-15-git-send-email-paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Inspection is proving insufficient to catch all RCU misuses,
which is understandable given that rcu_dereference() might be
protected by any of four different flavors of RCU (RCU, RCU-bh,
RCU-sched, and SRCU), and might also/instead be protected by any
of a number of locking primitives. It is therefore time to
enlist the aid of lockdep.
This set of patches is inspired by earlier work by Peter
Zijlstra and Thomas Gleixner, and takes the following approach:
o Set up separate lockdep classes for RCU, RCU-bh, and RCU-sched.
o Set up separate lockdep classes for each instance of SRCU.
o Create primitives that check for being in an RCU read-side
critical section. These return exact answers if lockdep is
fully enabled, but if unsure, report being in an RCU read-side
critical section. (We want to avoid false positives!)
The primitives are:
For RCU: rcu_read_lock_held(void)
For RCU-bh: rcu_read_lock_bh_held(void)
For RCU-sched: rcu_read_lock_sched_held(void)
For SRCU: srcu_read_lock_held(struct srcu_struct *sp)
o Add rcu_dereference_check(), which takes a second argument
in which one places a boolean expression based on the above
primitives and/or lockdep_is_held().
o A new kernel configuration parameter, CONFIG_PROVE_RCU, enables
rcu_dereference_check(). This depends on CONFIG_PROVE_LOCKING,
and should be quite helpful during the transition period while
CONFIG_PROVE_RCU-unaware patches are in flight.
The existing rcu_dereference() primitive does no checking, but
upcoming patches will change that.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: laijs@cn.fujitsu.com
Cc: dipankar@in.ibm.com
Cc: mathieu.desnoyers@polymtl.ca
Cc: josh@joshtriplett.org
Cc: dvhltc@us.ibm.com
Cc: niv@us.ibm.com
Cc: peterz@infradead.org
Cc: rostedt@goodmis.org
Cc: Valdis.Kletnieks@vt.edu
Cc: dhowells@redhat.com
LKML-Reference: <1266887105-1528-1-git-send-email-paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Return -EINVAL for the bad size and for unrecognized NT_* type in
ptrace_regset() instead of -EIO.
Also update the comments for this ptrace interface with more clarifications.
Requested-by: Roland McGrath <roland@redhat.com>
Requested-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
LKML-Reference: <20100222225240.397523600@sbs-t61.sc.intel.com>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
find_task_by_vpid() is not safe without rcu_read_lock(). 2.6.33-rc7 got
RCU protection for sys_setpriority() but missed it for sys_getpriority().
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Introduce run-time PM callbacks for the PCI bus type. Make the new
callbacks work in analogy with the existing system sleep PM
callbacks, so that the drivers already converted to struct dev_pm_ops
can use their suspend and resume routines for run-time PM without
modifications.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Useful for freeing a portion of the resource tree, e.g. when trying to
reallocate resources more efficiently.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Now that we return the new resource start position, there is no
need to update "struct resource" inside the align function.
Therefore, mark the struct resource as const.
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
As suggested by Linus, align functions should return the start
of a resource, not void. An update of "res->start" is no longer
necessary.
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
KPROBES_EVENT actually depends on the regs and stack access API
(b1cf540f) and not on x86.
So introduce a new config option which architectures can select if
they have the API implemented and switch x86.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
LKML-Reference: <20100210162517.GB6933@osiris.boeblingen.de.ibm.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Most implementations of arch_syscall_addr() are the same, so create a
default version in common code and move the one piece that differs (the
syscall table) to asm/syscall.h. New arch ports don't have to waste
time copying & pasting this simple function.
The s390/sparc versions need to be different, so document why.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Paul Mundt <lethal@linux-sh.org>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1264498803-17278-1-git-send-email-vapier@gentoo.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
setscheduler() saves task->sched_class outside of the rq->lock held
region for a check after the setscheduler changes have become
effective. That might result in checking a stale value.
rtmutex_setprio() has the same problem, though it is protected by
p->pi_lock against setscheduler(), but for correctness sake (and to
avoid bad examples) it needs to be fixed as well.
Retrieve task->sched_class inside of the rq->lock held region.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: stable@kernel.org
This patch fixes following sparse warnings:
include/linux/kfifo.h:127:25: warning: Using plain integer as NULL pointer
kernel/kfifo.c:83:21: warning: Using plain integer as NULL pointer
Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Acked-by: Stefani Seibold <stefani@seibold.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
After kfifo rework it's no longer possible to reliably know if kfifo is
usable, since after kfifo_free(), kfifo_initialized() would still return
true. The correct behaviour is needed for at least FHCI USB driver.
This patch fixes the issue by resetting the kfifo to zero values (the
same approach is used in kfifo_alloc() if allocation failed).
Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Acked-by: Stefani Seibold <stefani@seibold.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Thomas found that due to ttwu() changing a task's cpu without holding
the rq->lock, task_rq_lock() might end up locking the wrong rq.
Avoid this by serializing against TASK_WAKING.
Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1266241712.15770.420.camel@laptop>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Fix a SMT scheduler performance regression that is leading to a scenario
where SMT threads in one core are completely idle while both the SMT threads
in another core (on the same socket) are busy.
This is caused by this commit (with the problematic code highlighted)
commit bdb94aa5db
Author: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Tue Sep 1 10:34:38 2009 +0200
sched: Try to deal with low capacity
@@ -4203,15 +4223,18 @@ find_busiest_queue()
...
for_each_cpu(i, sched_group_cpus(group)) {
+ unsigned long power = power_of(i);
...
- wl = weighted_cpuload(i);
+ wl = weighted_cpuload(i) * SCHED_LOAD_SCALE;
+ wl /= power;
- if (rq->nr_running == 1 && wl > imbalance)
+ if (capacity && rq->nr_running == 1 && wl > imbalance)
continue;
On a SMT system, power of the HT logical cpu will be 589 and
the scheduler load imbalance (for scenarios like the one mentioned above)
can be approximately 1024 (SCHED_LOAD_SCALE). The above change of scaling
the weighted load with the power will result in "wl > imbalance" and
ultimately resulting in find_busiest_queue() return NULL, causing
load_balance() to think that the load is well balanced. But infact
one of the tasks can be moved to the idle core for optimal performance.
We don't need to use the weighted load (wl) scaled by the cpu power to
compare with imabalance. In that condition, we already know there is only a
single task "rq->nr_running == 1" and the comparison between imbalance,
wl is to make sure that we select the correct priority thread which matches
imbalance. So we really need to compare the imabalnce with the original
weighted load of the cpu and not the scaled load.
But in other conditions where we want the most hammered(busiest) cpu, we can
use scaled load to ensure that we consider the cpu power in addition to the
actual load on that cpu, so that we can move the load away from the
guy that is getting most hammered with respect to the actual capacity,
as compared with the rest of the cpu's in that busiest group.
Fix it.
Reported-by: Ma Ling <ling.ma@intel.com>
Initial-Analysis-by: Zhang, Yanmin <yanmin_zhang@linux.intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1266023662.2808.118.camel@sbs-t61.sc.intel.com>
Cc: stable@kernel.org [2.6.32.x]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Dmitry Monakhov