Fernando Lopez-Lezcano reported frequent scheduling latencies and audio
xruns starting at the 2.6.18-rt kernel, and those problems persisted all
until current -rt kernels. The latencies were serious and unjustified by
system load, often in the milliseconds range.
After a patient and heroic multi-month effort of Fernando, where he
tested dozens of kernels, tried various configs, boot options,
test-patches of mine and provided latency traces of those incidents, the
following 'smoking gun' trace was captured by him:
_------=> CPU#
/ _-----=> irqs-off
| / _----=> need-resched
|| / _---=> hardirq/softirq
||| / _--=> preempt-depth
|||| /
||||| delay
cmd pid ||||| time | caller
\ / ||||| \ | /
IRQ_19-1479 1D..1 0us : __trace_start_sched_wakeup (try_to_wake_up)
IRQ_19-1479 1D..1 0us : __trace_start_sched_wakeup <<...>-5856> (37 0)
IRQ_19-1479 1D..1 0us : __trace_start_sched_wakeup (c01262ba 0 0)
IRQ_19-1479 1D..1 0us : resched_task (try_to_wake_up)
IRQ_19-1479 1D..1 0us : __spin_unlock_irqrestore (try_to_wake_up)
...
<idle>-0 1...1 11us!: default_idle (cpu_idle)
...
<idle>-0 0Dn.1 602us : smp_apic_timer_interrupt (c0103baf 1 0)
...
<...>-5856 0D..2 618us : __switch_to (__schedule)
<...>-5856 0D..2 618us : __schedule <<idle>-0> (20 162)
<...>-5856 0D..2 619us : __spin_unlock_irq (__schedule)
<...>-5856 0...1 619us : trace_stop_sched_switched (__schedule)
<...>-5856 0D..1 619us : trace_stop_sched_switched <<...>-5856> (37 0)
what is visible in this trace is that CPU#1 ran try_to_wake_up() for
PID:5856, it placed PID:5856 on CPU#0's runqueue and ran resched_task()
for CPU#0. But it decided to not send an IPI that no CPU - due to
TS_POLLING. But CPU#0 never woke up after its NEED_RESCHED bit was set,
and only rescheduled to PID:5856 upon the next lapic timer IRQ. The
result was a 600+ usecs latency and a missed wakeup!
the bug turned out to be an idle-wakeup bug introduced into the mainline
kernel this summer via an optimization in the x86_64 tree:
commit 495ab9c045
Author: Andi Kleen <ak@suse.de>
Date: Mon Jun 26 13:59:11 2006 +0200
[PATCH] i386/x86-64/ia64: Move polling flag into thread_info_status
During some profiling I noticed that default_idle causes a lot of
memory traffic. I think that is caused by the atomic operations
to clear/set the polling flag in thread_info. There is actually
no reason to make this atomic - only the idle thread does it
to itself, other CPUs only read it. So I moved it into ti->status.
the problem is this type of change:
if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
- clear_thread_flag(TIF_POLLING_NRFLAG);
+ current_thread_info()->status &= ~TS_POLLING;
smp_mb__after_clear_bit();
while (!need_resched()) {
local_irq_disable();
this changes clear_thread_flag() to an explicit clearing of TS_POLLING.
clear_thread_flag() is defined as:
clear_bit(flag, &ti->flags);
and clear_bit() is a LOCK-ed atomic instruction on all x86 platforms:
static inline void clear_bit(int nr, volatile unsigned long * addr)
{
__asm__ __volatile__( LOCK_PREFIX
"btrl %1,%0"
hence smp_mb__after_clear_bit() is defined as a simple compile barrier:
#define smp_mb__after_clear_bit() barrier()
but the explicit TS_POLLING clearing introduced by the patch:
+ current_thread_info()->status &= ~TS_POLLING;
is not an atomic op! So the clearing of the TS_POLLING bit is freely
reorderable with the reading of the NEED_RESCHED bit - and both now
reside in different memory addresses.
CPU idle wakeup very much depends on ordered memory ops, the clearing of
the TS_POLLING flag must always be done before we test need_resched()
and hit the idle instruction(s). [Symmetrically, the wakeup code needs
to set NEED_RESCHED before it tests the TS_POLLING flag, so memory
ordering is paramount.]
Fernando's dual-core Athlon64 system has a sufficiently advanced memory
ordering model so that it triggered this scenario very often.
( And it also turned out that the reason why these latencies never
triggered on my testsystems is that i routinely use idle=poll, which
was the only idle variant not affected by this bug. )
The fix is to change the smp_mb__after_clear_bit() to an smp_mb(), to
act as an absolute barrier between the TS_POLLING write and the
NEED_RESCHED read. This affects almost all idling methods (default,
ACPI, APM), on all 3 x86 architectures: i386, x86_64, ia64.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Tested-by: Fernando Lopez-Lezcano <nando@ccrma.Stanford.EDU>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The last in-kernel user of errno is gone, so we should remove the definition
and everything referring to it. This also removes the now-unused lib/execve.c
file that was introduced earlier.
Also remove every trace of __KERNEL_SYSCALLS__ that still remained in the
kernel.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Andi Kleen <ak@muc.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ian Molton <spyro@f2s.com>
Cc: Mikael Starvik <starvik@axis.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Hirokazu Takata <takata.hirokazu@renesas.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp>
Cc: Richard Curnow <rc@rc0.org.uk>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp>
Cc: Chris Zankel <chris@zankel.net>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
During some profiling I noticed that default_idle causes a lot of
memory traffic. I think that is caused by the atomic operations
to clear/set the polling flag in thread_info. There is actually
no reason to make this atomic - only the idle thread does it
to itself, other CPUs only read it. So I moved it into ti->status.
Converted i386/x86-64/ia64 for now because that was the easiest
way to fix ACPI which also manipulates these flags in its idle
function.
Cc: Nick Piggin <npiggin@novell.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Len Brown <len.brown@intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When kretprobe probes the schedule() function, if the probed process exits
then schedule() will never return, so some kretprobe instances will never
be recycled.
In this patch the parent process will recycle retprobe instances of the
probed function and there will be no memory leak of kretprobe instances.
Signed-off-by: bibo mao <bibo.mao@intel.com>
Cc: Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
We have a customer application which trips a bug. The problem arises
when a driver attempts to call do_munmap on an area which is mapped, but
because current->thread.task_size has been set to 0xC0000000, the call
to do_munmap fails thinking it is an unmap beyond the user's address
space.
The comment in fs/binfmt_elf.c in load_elf_library() before the call
to SET_PERSONALITY() indicates that task_size must not be changed for
the running application until flush_thread, but is for ia64 executing
ia32 binaries.
This patch moves the setting of task_size from SET_PERSONALITY() to
flush_thread() as indicated. The customer application no longer is able
to trip the bug.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Fix a bug in kprobes that can cause an Oops or even a crash when a return
probe is installed on one of the following functions: sys_execve,
do_execve, load_*_binary, flush_old_exec, or flush_thread. The fix is to
remove the call to kprobe_flush_task() in flush_thread(). This fix has
been tested on all architectures for which the return-probes feature has
been implemented (i386, x86_64, ppc64, ia64). Please apply.
BACKGROUND
Up to now, we have called kprobe_flush_task() under two situations: when a
task exits, and when it execs. Flushing kretprobe_instances on exit is
correct because (a) do_exit() doesn't return, and (b) one or more
return-probed functions may be active when a task calls do_exit(). Neither
is the case for sys_execve() and its callees.
Initially, the mistaken call to kprobe_flush_task() on exec was harmless
because we put the "real" return address of each active probed function
back in the stack, just to be safe, when we recycled its
kretprobe_instance. When support for ppc64 and ia64 was added, this safety
measure couldn't be employed, and was eventually dropped even for i386 and
x86_64. sys_execve() and its callees were informally blacklisted for
return probes until this fix was developed.
Acked-by: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Our performance validation on 2.6.15-rc1 caught a disastrous performance
regression on ia64 with netperf (-98%) and volanomark (-58%) compares to
previous kernel version 2.6.14-git7. See the following chart (result
group 1 & 2).
http://kernel-perf.sourceforge.net/results.machine_id=26.html
We have root caused it to commit 64c7c8f885
This changeset broke the ia64 task resched notification. In
sched.c:resched_task(), a reschedule IPI is conditioned upon
TIF_POLLING_NRFLAG. However, the above changeset unconditionally set
the polling thread flag for idle tasks regardless whether pal_halt_light
is in use or not. As a result, resched IPI is not sent from
resched_task(). And since the default behavior on ia64 is to use
pal_halt_light, we end up delaying the rescheduling task until next
timer tick, and thus cause the performance regression.
This fixes the performance bug. I'm glad our performance suite is
turning up bad performance bug like this in time.
Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Make some changes to the NEED_RESCHED and POLLING_NRFLAG to reduce
confusion, and make their semantics rigid. Improves efficiency of
resched_task and some cpu_idle routines.
* In resched_task:
- TIF_NEED_RESCHED is only cleared with the task's runqueue lock held,
and as we hold it during resched_task, then there is no need for an
atomic test and set there. The only other time this should be set is
when the task's quantum expires, in the timer interrupt - this is
protected against because the rq lock is irq-safe.
- If TIF_NEED_RESCHED is set, then we don't need to do anything. It
won't get unset until the task get's schedule()d off.
- If we are running on the same CPU as the task we resched, then set
TIF_NEED_RESCHED and no further action is required.
- If we are running on another CPU, and TIF_POLLING_NRFLAG is *not* set
after TIF_NEED_RESCHED has been set, then we need to send an IPI.
Using these rules, we are able to remove the test and set operation in
resched_task, and make clear the previously vague semantics of
POLLING_NRFLAG.
* In idle routines:
- Enter cpu_idle with preempt disabled. When the need_resched() condition
becomes true, explicitly call schedule(). This makes things a bit clearer
(IMO), but haven't updated all architectures yet.
- Many do a test and clear of TIF_NEED_RESCHED for some reason. According
to the resched_task rules, this isn't needed (and actually breaks the
assumption that TIF_NEED_RESCHED is only cleared with the runqueue lock
held). So remove that. Generally one less locked memory op when switching
to the idle thread.
- Many idle routines clear TIF_POLLING_NRFLAG, and only set it in the inner
most polling idle loops. The above resched_task semantics allow it to be
set until before the last time need_resched() is checked before going into
a halt requiring interrupt wakeup.
Many idle routines simply never enter such a halt, and so POLLING_NRFLAG
can be always left set, completely eliminating resched IPIs when rescheduling
the idle task.
POLLING_NRFLAG width can be increased, to reduce the chance of resched IPIs.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Con Kolivas <kernel@kolivas.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Run idle threads with preempt disabled.
Also corrected a bugs in arm26's cpu_idle (make it actually call schedule()).
How did it ever work before?
Might fix the CPU hotplugging hang which Nigel Cunningham noted.
We think the bug hits if the idle thread is preempted after checking
need_resched() and before going to sleep, then the CPU offlined.
After calling stop_machine_run, the CPU eventually returns from preemption and
into the idle thread and goes to sleep. The CPU will continue executing
previous idle and have no chance to call play_dead.
By disabling preemption until we are ready to explicitly schedule, this bug is
fixed and the idle threads generally become more robust.
From: alexs <ashepard@u.washington.edu>
PPC build fix
From: Yoichi Yuasa <yuasa@hh.iij4u.or.jp>
MIPS build fix
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
notify_die() added for MCA_{MONARCH,SLAVE,RENDEZVOUS}_{ENTER,PROCESS,LEAVE} and
INIT_{MONARCH,SLAVE}_{ENTER,PROCESS,LEAVE}. We need multiple
notification points for these events because they can take many seconds
to run which has nasty effects on the behaviour of the rest of the
system.
DIE_SS replaced by a generic DIE_FAULT which checks the vector number,
to allow interception of faults other than SS.
DIE_MACHINE_{HALT,RESTART} added to allow last minute close down
processing, especially when the halt/restart routines are called from
error handlers.
DIE_OOPS added.
The check for kprobe's break numbers has been moved from traps.c to
kprobes.c, allowing DIE_BREAK to be used for any additional break
numbers, i.e. it is no longer kprobes specific.
Hooks for kernel debuggers and kernel dumpers added, ENTER and LEAVE.
Both of these disable the system for long periods which impact on
watchdogs and heartbeat systems in general. More patches to come that
use these events to reset watchdogs and heartbeats.
unregister_die_notifier() added and both routines exported. Requested
by Dean Nelson.
Lock removed from {un,}register_die_notifier. notifier_chain_register()
already takes a lock. Also the generic notifier chain locking is being
reworked to distinguish between callbacks that can block and those that
cannot, the lock in {un,}register_die_notifier would interfere with
that change. http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
Leading white space removed from arch/ia64/kernel/kprobes.c.
Typo in mca.c in original version of this patch found & fixed by Dean
Nelson.
Signed-off-by: Keith Owens <kaos@sgi.com>
Acked-by: Dean Nelson <dcn@sgi.com>
Acked-by: Anil Keshavamurthy <anil.s.keshavamurthy@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
this changeset broke the "nohalt" kernel boot option.
8df5a500a3
default_idle() is looking at new variable can_do_pal_halt. However,
that variable did not get cleared upon "nohalt" boot option. Result
is that "nohalt" option is ignored until perfmon is exercised.
Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
machine_restart, machine_halt and machine_power_off are machine
specific hooks deep into the reboot logic, that modules
have no business messing with. Usually code should be calling
kernel_restart, kernel_halt, kernel_power_off, or
emergency_restart. So don't export machine_restart,
machine_halt, and machine_power_off so we can catch buggy users.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The following patch implements function return probes for ia64 using
the revised design. With this new design we no longer need to do some
of the odd hacks previous required on the last ia64 return probe port
that I sent out for comments.
Note that this new implementation still does not resolve the problem noted
by Keith Owens where backtrace data is lost after a return probe is hit.
Changes include:
* Addition of kretprobe_trampoline to act as a dummy function for instrumented
functions to return to, and for the return probe infrastructure to place
a kprobe on on, gaining control so that the return probe handler
can be called, and so that the instruction pointer can be moved back
to the original return address.
* Addition of arch_init(), allowing a kprobe to be registered on
kretprobe_trampoline
* Addition of trampoline_probe_handler() which is used as the pre_handler
for the kprobe inserted on kretprobe_implementation. This is the function
that handles the details for calling the return probe handler function
and returning control back at the original return address
* Addition of arch_prepare_kretprobe() which is setup as the pre_handler
for a kprobe registered at the beginning of the target function by
kernel/kprobes.c so that a return probe instance can be setup when
a caller enters the target function. (A return probe instance contains
all the needed information for trampoline_probe_handler to do it's job.)
* Hooks added to the exit path of a task so that we can cleanup any left-over
return probe instances (i.e. if a task dies while inside a targeted function
then the return probe instance was reserved at the beginning of the function
but the function never returns so we need to mark the instance as unused.)
Signed-off-by: Rusty Lynch <rusty.lynch@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The pmu_active test is based on the values of PSR.up. THIS IS THE PROBLEM as
it does not take into account the lazy restore logic which is as follow (simplified):
context switch out:
save PMDs
clear psr.up
release ownership
context switch in:
if (ctx->last_cpu == smp_processor_id() && ctx->cpu_activation == cpu_activation) {
set psr.up
return
}
restore PMD
restore PMC
ctx->last_cpu = smp_processor_id();
ctx->activation = ++cpu_activation;
set psr.up
The key here is that on context switch out, we clear psr.up and on context switch in
we check if nobody else used the PMU on that processor since last time we came. In
that case, we assume the PMD/PMC are ours and we simply reactivate.
The Caliper problem is that between the moment we context switch out and the moment we
come back, nobody effectively used the PMU BUT the processor went idle. Normally this
would have no incidence but PAL_HALT does alter the PMU registers. In default_idle(),
the test on psr.up is not strong enough to cover this case and we go into PAL which
trashed the PMU resgisters. When we come back we falsely assume that this is our state
yet it is corrupted. Very nasty indeed.
To avoid the problem it is necessary to forbid going to PAL_HALT as soon as perfmon
installs some valid state in the PMU registers. This happens with an application
attaches a context to a thread or CPU. It is not enough to check the psr/dcr bits.
Hence I propose the attached patch. It adds a callback in process.c to modify the
condition to enter PAL on idle. Basically, now it is conditional to pal_halt=1 AND
perfmon saying it is okay.
Signed-off-by: Tony Luck <tony.luck@intel.com>
Andi noted that during normal runtime cpu_idle_map is bounced around a lot,
and occassionally at a higher frequency than the timer interrupt wakeup
which we normally exit pm_idle from. So switch to a percpu variable.
I didn't move things to the slow path because it would involve adding
scheduler code to wakeup the idle thread on the cpus we're waiting for.
Signed-off-by: Zwane Mwaikambo <zwane@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This patch is required to support cpu removal for IPF systems. Existing code
just fakes the real offline by keeping it run the idle thread, and polling
for the bit to re-appear in the cpu_state to get out of the idle loop.
For the cpu-offline to work correctly, we need to pass control of this CPU
back to SAL so it can continue in the boot-rendez mode. This gives the
SAL control to not pick this cpu as the monarch processor for global MCA
events, and addition does not wait for this cpu to checkin with SAL
for global MCA events as well. The handoff is implemented as documented in
SAL specification section 3.2.5.1 "OS_BOOT_RENDEZ to SAL return State"
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
