Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
/*-
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
* Copyright (c) 2010-2013 Alexander Motin <mav@FreeBSD.org>
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer,
|
|
|
|
* without modification, immediately at the beginning of the file.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
|
|
|
|
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
|
|
|
|
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
|
|
|
|
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
|
|
|
|
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
|
|
|
|
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
|
|
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
|
|
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
|
|
|
|
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Common routines to manage event timers hardware.
|
|
|
|
*/
|
|
|
|
|
2010-09-22 05:32:37 +00:00
|
|
|
#include "opt_device_polling.h"
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
|
|
|
#include <sys/param.h>
|
|
|
|
#include <sys/systm.h>
|
|
|
|
#include <sys/bus.h>
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
#include <sys/limits.h>
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
#include <sys/lock.h>
|
|
|
|
#include <sys/kdb.h>
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
#include <sys/ktr.h>
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
#include <sys/mutex.h>
|
|
|
|
#include <sys/proc.h>
|
|
|
|
#include <sys/kernel.h>
|
|
|
|
#include <sys/sched.h>
|
|
|
|
#include <sys/smp.h>
|
|
|
|
#include <sys/sysctl.h>
|
|
|
|
#include <sys/timeet.h>
|
2010-09-14 08:48:06 +00:00
|
|
|
#include <sys/timetc.h>
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
|
|
|
#include <machine/atomic.h>
|
|
|
|
#include <machine/clock.h>
|
|
|
|
#include <machine/cpu.h>
|
|
|
|
#include <machine/smp.h>
|
|
|
|
|
2015-01-05 20:44:44 +00:00
|
|
|
int cpu_disable_c2_sleep = 0; /* Timer dies in C2. */
|
|
|
|
int cpu_disable_c3_sleep = 0; /* Timer dies in C3. */
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
static void setuptimer(void);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
static void loadtimer(sbintime_t now, int first);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
static int doconfigtimer(void);
|
|
|
|
static void configtimer(int start);
|
|
|
|
static int round_freq(struct eventtimer *et, int freq);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
static sbintime_t getnextcpuevent(int idle);
|
|
|
|
static sbintime_t getnextevent(void);
|
|
|
|
static int handleevents(sbintime_t now, int fake);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
static struct mtx et_hw_mtx;
|
|
|
|
|
|
|
|
#define ET_HW_LOCK(state) \
|
|
|
|
{ \
|
|
|
|
if (timer->et_flags & ET_FLAGS_PERCPU) \
|
|
|
|
mtx_lock_spin(&(state)->et_hw_mtx); \
|
|
|
|
else \
|
|
|
|
mtx_lock_spin(&et_hw_mtx); \
|
|
|
|
}
|
|
|
|
|
|
|
|
#define ET_HW_UNLOCK(state) \
|
|
|
|
{ \
|
|
|
|
if (timer->et_flags & ET_FLAGS_PERCPU) \
|
|
|
|
mtx_unlock_spin(&(state)->et_hw_mtx); \
|
|
|
|
else \
|
|
|
|
mtx_unlock_spin(&et_hw_mtx); \
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct eventtimer *timer = NULL;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
static sbintime_t timerperiod; /* Timer period for periodic mode. */
|
|
|
|
static sbintime_t statperiod; /* statclock() events period. */
|
|
|
|
static sbintime_t profperiod; /* profclock() events period. */
|
|
|
|
static sbintime_t nexttick; /* Next global timer tick time. */
|
|
|
|
static u_int busy = 1; /* Reconfiguration is in progress. */
|
2014-06-28 03:56:17 +00:00
|
|
|
static int profiling; /* Profiling events enabled. */
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
static char timername[32]; /* Wanted timer. */
|
|
|
|
TUNABLE_STR("kern.eventtimer.timer", timername, sizeof(timername));
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
2014-06-28 03:56:17 +00:00
|
|
|
static int singlemul; /* Multiplier for periodic mode. */
|
|
|
|
SYSCTL_INT(_kern_eventtimer, OID_AUTO, singlemul, CTLFLAG_RWTUN, &singlemul,
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
0, "Multiplier for periodic mode");
|
|
|
|
|
2014-06-28 03:56:17 +00:00
|
|
|
static u_int idletick; /* Run periodic events when idle. */
|
|
|
|
SYSCTL_UINT(_kern_eventtimer, OID_AUTO, idletick, CTLFLAG_RWTUN, &idletick,
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
0, "Run periodic events when idle");
|
|
|
|
|
2014-06-28 03:56:17 +00:00
|
|
|
static int periodic; /* Periodic or one-shot mode. */
|
|
|
|
static int want_periodic; /* What mode to prefer. */
|
2010-09-21 16:50:24 +00:00
|
|
|
TUNABLE_INT("kern.eventtimer.periodic", &want_periodic);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
struct pcpu_state {
|
|
|
|
struct mtx et_hw_mtx; /* Per-CPU timer mutex. */
|
|
|
|
u_int action; /* Reconfiguration requests. */
|
|
|
|
u_int handle; /* Immediate handle resuests. */
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now; /* Last tick time. */
|
|
|
|
sbintime_t nextevent; /* Next scheduled event on this CPU. */
|
|
|
|
sbintime_t nexttick; /* Next timer tick time. */
|
2015-07-20 09:37:42 +00:00
|
|
|
sbintime_t nexthard; /* Next hardclock() event. */
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t nextstat; /* Next statclock() event. */
|
|
|
|
sbintime_t nextprof; /* Next profclock() event. */
|
|
|
|
sbintime_t nextcall; /* Next callout event. */
|
|
|
|
sbintime_t nextcallopt; /* Next optional callout event. */
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
int ipi; /* This CPU needs IPI. */
|
|
|
|
int idle; /* This CPU is in idle mode. */
|
|
|
|
};
|
|
|
|
|
2010-11-22 19:32:54 +00:00
|
|
|
static DPCPU_DEFINE(struct pcpu_state, timerstate);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
DPCPU_DEFINE(sbintime_t, hardclocktime);
|
2013-02-28 13:46:03 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Timer broadcast IPI handler.
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
hardclockintr(void)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
|
|
|
int done;
|
|
|
|
|
|
|
|
if (doconfigtimer() || busy)
|
|
|
|
return (FILTER_HANDLED);
|
|
|
|
state = DPCPU_PTR(timerstate);
|
|
|
|
now = state->now;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
CTR3(KTR_SPARE2, "ipi at %d: now %d.%08x",
|
|
|
|
curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
|
|
|
|
done = handleevents(now, 0);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
return (done ? FILTER_HANDLED : FILTER_STRAY);
|
|
|
|
}
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Handle all events for specified time on this CPU
|
|
|
|
*/
|
|
|
|
static int
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
handleevents(sbintime_t now, int fake)
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t t, *hct;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct trapframe *frame;
|
|
|
|
struct pcpu_state *state;
|
|
|
|
int usermode;
|
|
|
|
int done, runs;
|
|
|
|
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
CTR3(KTR_SPARE2, "handle at %d: now %d.%08x",
|
|
|
|
curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
done = 0;
|
|
|
|
if (fake) {
|
|
|
|
frame = NULL;
|
|
|
|
usermode = 0;
|
|
|
|
} else {
|
|
|
|
frame = curthread->td_intr_frame;
|
|
|
|
usermode = TRAPF_USERMODE(frame);
|
|
|
|
}
|
2011-05-16 15:29:59 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state = DPCPU_PTR(timerstate);
|
2011-05-16 15:29:59 +00:00
|
|
|
|
2012-03-10 14:57:21 +00:00
|
|
|
runs = 0;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
while (now >= state->nexthard) {
|
|
|
|
state->nexthard += tick_sbt;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
runs++;
|
|
|
|
}
|
2012-08-03 09:08:20 +00:00
|
|
|
if (runs) {
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
hct = DPCPU_PTR(hardclocktime);
|
|
|
|
*hct = state->nexthard - tick_sbt;
|
2012-08-03 09:08:20 +00:00
|
|
|
if (fake < 2) {
|
|
|
|
hardclock_cnt(runs, usermode);
|
|
|
|
done = 1;
|
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
2012-03-10 14:57:21 +00:00
|
|
|
runs = 0;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
while (now >= state->nextstat) {
|
|
|
|
state->nextstat += statperiod;
|
2012-03-10 14:57:21 +00:00
|
|
|
runs++;
|
|
|
|
}
|
|
|
|
if (runs && fake < 2) {
|
|
|
|
statclock_cnt(runs, usermode);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
done = 1;
|
|
|
|
}
|
|
|
|
if (profiling) {
|
2012-03-10 14:57:21 +00:00
|
|
|
runs = 0;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
while (now >= state->nextprof) {
|
|
|
|
state->nextprof += profperiod;
|
2012-03-10 14:57:21 +00:00
|
|
|
runs++;
|
|
|
|
}
|
|
|
|
if (runs && !fake) {
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
profclock_cnt(runs, usermode, TRAPF_PC(frame));
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
done = 1;
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
state->nextprof = state->nextstat;
|
2017-01-20 17:40:31 +00:00
|
|
|
if (now >= state->nextcallopt || now >= state->nextcall) {
|
2014-04-12 23:29:29 +00:00
|
|
|
state->nextcall = state->nextcallopt = SBT_MAX;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
callout_process(now);
|
|
|
|
}
|
2011-05-16 15:29:59 +00:00
|
|
|
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
t = getnextcpuevent(0);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_HW_LOCK(state);
|
|
|
|
if (!busy) {
|
|
|
|
state->idle = 0;
|
|
|
|
state->nextevent = t;
|
2013-12-16 13:52:18 +00:00
|
|
|
loadtimer(now, (fake == 2) &&
|
|
|
|
(timer->et_flags & ET_FLAGS_PERCPU));
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
|
|
|
ET_HW_UNLOCK(state);
|
|
|
|
return (done);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Schedule binuptime of the next event on current CPU.
|
|
|
|
*/
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
static sbintime_t
|
|
|
|
getnextcpuevent(int idle)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t event;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
u_int hardfreq;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
state = DPCPU_PTR(timerstate);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
/* Handle hardclock() events, skipping some if CPU is idle. */
|
|
|
|
event = state->nexthard;
|
|
|
|
if (idle) {
|
|
|
|
hardfreq = (u_int)hz / 2;
|
|
|
|
if (tc_min_ticktock_freq > 2
|
|
|
|
#ifdef SMP
|
|
|
|
&& curcpu == CPU_FIRST()
|
|
|
|
#endif
|
|
|
|
)
|
|
|
|
hardfreq = hz / tc_min_ticktock_freq;
|
|
|
|
if (hardfreq > 1)
|
|
|
|
event += tick_sbt * (hardfreq - 1);
|
|
|
|
}
|
|
|
|
/* Handle callout events. */
|
|
|
|
if (event > state->nextcall)
|
|
|
|
event = state->nextcall;
|
2012-03-13 10:21:08 +00:00
|
|
|
if (!idle) { /* If CPU is active - handle other types of events. */
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
if (event > state->nextstat)
|
|
|
|
event = state->nextstat;
|
|
|
|
if (profiling && event > state->nextprof)
|
|
|
|
event = state->nextprof;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
return (event);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Schedule binuptime of the next event on all CPUs.
|
|
|
|
*/
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
static sbintime_t
|
|
|
|
getnextevent(void)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t event;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
#ifdef SMP
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
int cpu;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
#endif
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
int c;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
state = DPCPU_PTR(timerstate);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
event = state->nextevent;
|
|
|
|
c = -1;
|
2012-03-13 10:21:08 +00:00
|
|
|
#ifdef SMP
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
if ((timer->et_flags & ET_FLAGS_PERCPU) == 0) {
|
|
|
|
CPU_FOREACH(cpu) {
|
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
if (event > state->nextevent) {
|
|
|
|
event = state->nextevent;
|
|
|
|
c = cpu;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
}
|
2012-03-13 10:21:08 +00:00
|
|
|
}
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
#endif
|
|
|
|
CTR4(KTR_SPARE2, "next at %d: next %d.%08x by %d",
|
|
|
|
curcpu, (int)(event >> 32), (u_int)(event & 0xffffffff), c);
|
|
|
|
return (event);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/* Hardware timer callback function. */
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
static void
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
timercb(struct eventtimer *et, void *arg)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now;
|
|
|
|
sbintime_t *next;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
|
|
|
#ifdef SMP
|
|
|
|
int cpu, bcast;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* Do not touch anything if somebody reconfiguring timers. */
|
|
|
|
if (busy)
|
|
|
|
return;
|
|
|
|
/* Update present and next tick times. */
|
|
|
|
state = DPCPU_PTR(timerstate);
|
|
|
|
if (et->et_flags & ET_FLAGS_PERCPU) {
|
|
|
|
next = &state->nexttick;
|
|
|
|
} else
|
|
|
|
next = &nexttick;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
now = sbinuptime();
|
|
|
|
if (periodic)
|
|
|
|
*next = now + timerperiod;
|
|
|
|
else
|
|
|
|
*next = -1; /* Next tick is not scheduled yet. */
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state->now = now;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
CTR3(KTR_SPARE2, "intr at %d: now %d.%08x",
|
|
|
|
curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
|
|
|
#ifdef SMP
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#ifdef EARLY_AP_STARTUP
|
|
|
|
MPASS(mp_ncpus == 1 || smp_started);
|
|
|
|
#endif
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/* Prepare broadcasting to other CPUs for non-per-CPU timers. */
|
|
|
|
bcast = 0;
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#ifdef EARLY_AP_STARTUP
|
|
|
|
if ((et->et_flags & ET_FLAGS_PERCPU) == 0) {
|
|
|
|
#else
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if ((et->et_flags & ET_FLAGS_PERCPU) == 0 && smp_started) {
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#endif
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
CPU_FOREACH(cpu) {
|
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
ET_HW_LOCK(state);
|
|
|
|
state->now = now;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
if (now >= state->nextevent) {
|
|
|
|
state->nextevent += SBT_1S;
|
2010-09-18 07:18:30 +00:00
|
|
|
if (curcpu != cpu) {
|
|
|
|
state->ipi = 1;
|
|
|
|
bcast = 1;
|
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
|
|
|
ET_HW_UNLOCK(state);
|
|
|
|
}
|
|
|
|
}
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
#endif
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
/* Handle events for this time on this CPU. */
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
handleevents(now, 0);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
#ifdef SMP
|
|
|
|
/* Broadcast interrupt to other CPUs for non-per-CPU timers. */
|
|
|
|
if (bcast) {
|
|
|
|
CPU_FOREACH(cpu) {
|
|
|
|
if (curcpu == cpu)
|
|
|
|
continue;
|
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
if (state->ipi) {
|
|
|
|
state->ipi = 0;
|
|
|
|
ipi_cpu(cpu, IPI_HARDCLOCK);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
#endif
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
* Load new value into hardware timer.
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
*/
|
|
|
|
static void
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
loadtimer(sbintime_t now, int start)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t new;
|
|
|
|
sbintime_t *next;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
uint64_t tmp;
|
|
|
|
int eq;
|
|
|
|
|
2010-11-08 15:25:12 +00:00
|
|
|
if (timer->et_flags & ET_FLAGS_PERCPU) {
|
|
|
|
state = DPCPU_PTR(timerstate);
|
|
|
|
next = &state->nexttick;
|
|
|
|
} else
|
|
|
|
next = &nexttick;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (periodic) {
|
|
|
|
if (start) {
|
|
|
|
/*
|
|
|
|
* Try to start all periodic timers aligned
|
|
|
|
* to period to make events synchronous.
|
|
|
|
*/
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
tmp = now % timerperiod;
|
|
|
|
new = timerperiod - tmp;
|
|
|
|
if (new < tmp) /* Left less then passed. */
|
|
|
|
new += timerperiod;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
CTR5(KTR_SPARE2, "load p at %d: now %d.%08x first in %d.%08x",
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff),
|
|
|
|
(int)(new >> 32), (u_int)(new & 0xffffffff));
|
|
|
|
*next = new + now;
|
|
|
|
et_start(timer, new, timerperiod);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
|
|
|
} else {
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
new = getnextevent();
|
|
|
|
eq = (new == *next);
|
|
|
|
CTR4(KTR_SPARE2, "load at %d: next %d.%08x eq %d",
|
|
|
|
curcpu, (int)(new >> 32), (u_int)(new & 0xffffffff), eq);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (!eq) {
|
|
|
|
*next = new;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
et_start(timer, new - now, 0);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
* Prepare event timer parameters after configuration changes.
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
*/
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
static void
|
|
|
|
setuptimer(void)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
int freq;
|
|
|
|
|
|
|
|
if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
|
|
|
|
periodic = 0;
|
|
|
|
else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
|
|
|
|
periodic = 1;
|
2010-09-14 04:48:04 +00:00
|
|
|
singlemul = MIN(MAX(singlemul, 1), 20);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
freq = hz * singlemul;
|
|
|
|
while (freq < (profiling ? profhz : stathz))
|
|
|
|
freq += hz;
|
|
|
|
freq = round_freq(timer, freq);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
timerperiod = SBT_1S / freq;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Reconfigure specified per-CPU timer on other CPU. Called from IPI handler.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
doconfigtimer(void)
|
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
|
|
|
|
|
|
|
state = DPCPU_PTR(timerstate);
|
|
|
|
switch (atomic_load_acq_int(&state->action)) {
|
|
|
|
case 1:
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
now = sbinuptime();
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_HW_LOCK(state);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
loadtimer(now, 1);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_HW_UNLOCK(state);
|
|
|
|
state->handle = 0;
|
|
|
|
atomic_store_rel_int(&state->action, 0);
|
|
|
|
return (1);
|
|
|
|
case 2:
|
|
|
|
ET_HW_LOCK(state);
|
|
|
|
et_stop(timer);
|
|
|
|
ET_HW_UNLOCK(state);
|
|
|
|
state->handle = 0;
|
|
|
|
atomic_store_rel_int(&state->action, 0);
|
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
if (atomic_readandclear_int(&state->handle) && !busy) {
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
now = sbinuptime();
|
|
|
|
handleevents(now, 0);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Reconfigure specified timer.
|
|
|
|
* For per-CPU timers use IPI to make other CPUs to reconfigure.
|
|
|
|
*/
|
|
|
|
static void
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
configtimer(int start)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now, next;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
int cpu;
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (start) {
|
|
|
|
setuptimer();
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
now = sbinuptime();
|
|
|
|
} else
|
|
|
|
now = 0;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
critical_enter();
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_HW_LOCK(DPCPU_PTR(timerstate));
|
|
|
|
if (start) {
|
|
|
|
/* Initialize time machine parameters. */
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
next = now + timerperiod;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (periodic)
|
|
|
|
nexttick = next;
|
|
|
|
else
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
nexttick = -1;
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#ifdef EARLY_AP_STARTUP
|
|
|
|
MPASS(mp_ncpus == 1 || smp_started);
|
|
|
|
#endif
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
CPU_FOREACH(cpu) {
|
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
state->now = now;
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#ifndef EARLY_AP_STARTUP
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
if (!smp_started && cpu != CPU_FIRST())
|
2014-04-12 23:29:29 +00:00
|
|
|
state->nextevent = SBT_MAX;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
else
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#endif
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
state->nextevent = next;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (periodic)
|
|
|
|
state->nexttick = next;
|
|
|
|
else
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
state->nexttick = -1;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state->nexthard = next;
|
|
|
|
state->nextstat = next;
|
|
|
|
state->nextprof = next;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
state->nextcall = next;
|
|
|
|
state->nextcallopt = next;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
hardclock_sync(cpu);
|
|
|
|
}
|
|
|
|
busy = 0;
|
|
|
|
/* Start global timer or per-CPU timer of this CPU. */
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
loadtimer(now, 1);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
} else {
|
|
|
|
busy = 1;
|
|
|
|
/* Stop global timer or per-CPU timer of this CPU. */
|
|
|
|
et_stop(timer);
|
|
|
|
}
|
|
|
|
ET_HW_UNLOCK(DPCPU_PTR(timerstate));
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
#ifdef SMP
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#ifdef EARLY_AP_STARTUP
|
|
|
|
/* If timer is global we are done. */
|
|
|
|
if ((timer->et_flags & ET_FLAGS_PERCPU) == 0) {
|
|
|
|
#else
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/* If timer is global or there is no other CPUs yet - we are done. */
|
|
|
|
if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || !smp_started) {
|
Add an EARLY_AP_STARTUP option to start APs earlier during boot.
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
|
|
|
#endif
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
critical_exit();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
/* Set reconfigure flags for other CPUs. */
|
|
|
|
CPU_FOREACH(cpu) {
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
atomic_store_rel_int(&state->action,
|
|
|
|
(cpu == curcpu) ? 0 : ( start ? 1 : 2));
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/* Broadcast reconfigure IPI. */
|
|
|
|
ipi_all_but_self(IPI_HARDCLOCK);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
/* Wait for reconfiguration completed. */
|
|
|
|
restart:
|
|
|
|
cpu_spinwait();
|
|
|
|
CPU_FOREACH(cpu) {
|
|
|
|
if (cpu == curcpu)
|
|
|
|
continue;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
if (atomic_load_acq_int(&state->action))
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
goto restart;
|
|
|
|
}
|
|
|
|
#endif
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
critical_exit();
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Calculate nearest frequency supported by hardware timer.
|
|
|
|
*/
|
2010-07-20 10:58:56 +00:00
|
|
|
static int
|
|
|
|
round_freq(struct eventtimer *et, int freq)
|
|
|
|
{
|
|
|
|
uint64_t div;
|
|
|
|
|
|
|
|
if (et->et_frequency != 0) {
|
2010-07-20 15:48:29 +00:00
|
|
|
div = lmax((et->et_frequency + freq / 2) / freq, 1);
|
2010-07-20 10:58:56 +00:00
|
|
|
if (et->et_flags & ET_FLAGS_POW2DIV)
|
|
|
|
div = 1 << (flsl(div + div / 2) - 1);
|
|
|
|
freq = (et->et_frequency + div / 2) / div;
|
|
|
|
}
|
2013-02-28 13:46:03 +00:00
|
|
|
if (et->et_min_period > SBT_1S)
|
2012-10-10 19:46:46 +00:00
|
|
|
panic("Event timer \"%s\" doesn't support sub-second periods!",
|
|
|
|
et->et_name);
|
2013-02-28 13:46:03 +00:00
|
|
|
else if (et->et_min_period != 0)
|
|
|
|
freq = min(freq, SBT2FREQ(et->et_min_period));
|
|
|
|
if (et->et_max_period < SBT_1S && et->et_max_period != 0)
|
|
|
|
freq = max(freq, SBT2FREQ(et->et_max_period));
|
2010-07-20 10:58:56 +00:00
|
|
|
return (freq);
|
|
|
|
}
|
|
|
|
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
/*
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
* Configure and start event timers (BSP part).
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
*/
|
|
|
|
void
|
|
|
|
cpu_initclocks_bsp(void)
|
|
|
|
{
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
|
|
|
int base, div, cpu;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
mtx_init(&et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
|
|
|
|
CPU_FOREACH(cpu) {
|
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
mtx_init(&state->et_hw_mtx, "et_hw_mtx", NULL, MTX_SPIN);
|
2014-04-12 23:29:29 +00:00
|
|
|
state->nextcall = SBT_MAX;
|
|
|
|
state->nextcallopt = SBT_MAX;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
2010-09-21 16:50:24 +00:00
|
|
|
periodic = want_periodic;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/* Grab requested timer or the best of present. */
|
|
|
|
if (timername[0])
|
|
|
|
timer = et_find(timername, 0, 0);
|
|
|
|
if (timer == NULL && periodic) {
|
|
|
|
timer = et_find(NULL,
|
|
|
|
ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
|
|
|
|
}
|
|
|
|
if (timer == NULL) {
|
|
|
|
timer = et_find(NULL,
|
|
|
|
ET_FLAGS_ONESHOT, ET_FLAGS_ONESHOT);
|
|
|
|
}
|
|
|
|
if (timer == NULL && !periodic) {
|
|
|
|
timer = et_find(NULL,
|
|
|
|
ET_FLAGS_PERIODIC, ET_FLAGS_PERIODIC);
|
|
|
|
}
|
|
|
|
if (timer == NULL)
|
2010-07-11 17:08:37 +00:00
|
|
|
panic("No usable event timer found!");
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
et_init(timer, timercb, NULL, NULL);
|
|
|
|
|
|
|
|
/* Adapt to timer capabilities. */
|
|
|
|
if (periodic && (timer->et_flags & ET_FLAGS_PERIODIC) == 0)
|
|
|
|
periodic = 0;
|
|
|
|
else if (!periodic && (timer->et_flags & ET_FLAGS_ONESHOT) == 0)
|
|
|
|
periodic = 1;
|
|
|
|
if (timer->et_flags & ET_FLAGS_C3STOP)
|
2015-01-05 20:44:44 +00:00
|
|
|
cpu_disable_c3_sleep++;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
/*
|
|
|
|
* We honor the requested 'hz' value.
|
|
|
|
* We want to run stathz in the neighborhood of 128hz.
|
|
|
|
* We would like profhz to run as often as possible.
|
|
|
|
*/
|
2010-09-14 04:48:04 +00:00
|
|
|
if (singlemul <= 0 || singlemul > 20) {
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
if (hz >= 1500 || (hz % 128) == 0)
|
|
|
|
singlemul = 1;
|
|
|
|
else if (hz >= 750)
|
|
|
|
singlemul = 2;
|
|
|
|
else
|
|
|
|
singlemul = 4;
|
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (periodic) {
|
|
|
|
base = round_freq(timer, hz * singlemul);
|
2010-07-20 10:58:56 +00:00
|
|
|
singlemul = max((base + hz / 2) / hz, 1);
|
|
|
|
hz = (base + singlemul / 2) / singlemul;
|
|
|
|
if (base <= 128)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
stathz = base;
|
|
|
|
else {
|
|
|
|
div = base / 128;
|
2010-07-20 10:58:56 +00:00
|
|
|
if (div >= singlemul && (div % singlemul) == 0)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
div++;
|
|
|
|
stathz = base / div;
|
|
|
|
}
|
|
|
|
profhz = stathz;
|
2010-07-20 10:58:56 +00:00
|
|
|
while ((profhz + stathz) <= 128 * 64)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
profhz += stathz;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
profhz = round_freq(timer, profhz);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
} else {
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
hz = round_freq(timer, hz);
|
|
|
|
stathz = round_freq(timer, 127);
|
|
|
|
profhz = round_freq(timer, stathz * 64);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
2010-07-20 15:48:29 +00:00
|
|
|
tick = 1000000 / hz;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
tick_sbt = SBT_1S / hz;
|
|
|
|
tick_bt = sbttobt(tick_sbt);
|
|
|
|
statperiod = SBT_1S / stathz;
|
|
|
|
profperiod = SBT_1S / profhz;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
ET_LOCK();
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
configtimer(1);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
ET_UNLOCK();
|
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Start per-CPU event timers on APs.
|
|
|
|
*/
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
void
|
|
|
|
cpu_initclocks_ap(void)
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
struct thread *td;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
2010-11-08 15:25:12 +00:00
|
|
|
state = DPCPU_PTR(timerstate);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
now = sbinuptime();
|
2010-11-08 15:25:12 +00:00
|
|
|
ET_HW_LOCK(state);
|
2012-08-04 08:06:37 +00:00
|
|
|
state->now = now;
|
2010-11-08 15:25:12 +00:00
|
|
|
hardclock_sync(curcpu);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
spinlock_enter();
|
2010-11-08 15:25:12 +00:00
|
|
|
ET_HW_UNLOCK(state);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
td = curthread;
|
|
|
|
td->td_intr_nesting_level++;
|
|
|
|
handleevents(state->now, 2);
|
|
|
|
td->td_intr_nesting_level--;
|
|
|
|
spinlock_exit();
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Switch to profiling clock rates.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
cpu_startprofclock(void)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
|
|
|
|
|
|
|
ET_LOCK();
|
2013-02-26 18:13:42 +00:00
|
|
|
if (profiling == 0) {
|
|
|
|
if (periodic) {
|
|
|
|
configtimer(0);
|
|
|
|
profiling = 1;
|
|
|
|
configtimer(1);
|
|
|
|
} else
|
|
|
|
profiling = 1;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
} else
|
2013-02-26 18:13:42 +00:00
|
|
|
profiling++;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
ET_UNLOCK();
|
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Switch to regular clock rates.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
cpu_stopprofclock(void)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_LOCK();
|
2013-02-26 18:13:42 +00:00
|
|
|
if (profiling == 1) {
|
|
|
|
if (periodic) {
|
|
|
|
configtimer(0);
|
|
|
|
profiling = 0;
|
|
|
|
configtimer(1);
|
|
|
|
} else
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
profiling = 0;
|
|
|
|
} else
|
2013-02-26 18:13:42 +00:00
|
|
|
profiling--;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_UNLOCK();
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Switch to idle mode (all ticks handled).
|
|
|
|
*/
|
2013-02-28 10:46:54 +00:00
|
|
|
sbintime_t
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
cpu_idleclock(void)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now, t;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (idletick || busy ||
|
2010-09-22 05:32:37 +00:00
|
|
|
(periodic && (timer->et_flags & ET_FLAGS_PERCPU))
|
|
|
|
#ifdef DEVICE_POLLING
|
|
|
|
|| curcpu == CPU_FIRST()
|
|
|
|
#endif
|
|
|
|
)
|
2013-02-28 10:46:54 +00:00
|
|
|
return (-1);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state = DPCPU_PTR(timerstate);
|
|
|
|
if (periodic)
|
|
|
|
now = state->now;
|
|
|
|
else
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
now = sbinuptime();
|
|
|
|
CTR3(KTR_SPARE2, "idle at %d: now %d.%08x",
|
|
|
|
curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
|
|
|
|
t = getnextcpuevent(1);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_HW_LOCK(state);
|
|
|
|
state->idle = 1;
|
|
|
|
state->nextevent = t;
|
|
|
|
if (!periodic)
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
loadtimer(now, 0);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_HW_UNLOCK(state);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
return (MAX(t - now, 0));
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Switch to active mode (skip empty ticks).
|
|
|
|
*/
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
void
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
cpu_activeclock(void)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
sbintime_t now;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
struct pcpu_state *state;
|
|
|
|
struct thread *td;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state = DPCPU_PTR(timerstate);
|
|
|
|
if (state->idle == 0 || busy)
|
|
|
|
return;
|
|
|
|
if (periodic)
|
|
|
|
now = state->now;
|
|
|
|
else
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
now = sbinuptime();
|
|
|
|
CTR3(KTR_SPARE2, "active at %d: now %d.%08x",
|
|
|
|
curcpu, (int)(now >> 32), (u_int)(now & 0xffffffff));
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
spinlock_enter();
|
|
|
|
td = curthread;
|
|
|
|
td->td_intr_nesting_level++;
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
handleevents(now, 1);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
td->td_intr_nesting_level--;
|
|
|
|
spinlock_exit();
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
}
|
|
|
|
|
2014-04-02 15:56:11 +00:00
|
|
|
/*
|
|
|
|
* Change the frequency of the given timer. This changes et->et_frequency and
|
|
|
|
* if et is the active timer it reconfigures the timer on all CPUs. This is
|
|
|
|
* intended to be a private interface for the use of et_change_frequency() only.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
cpu_et_frequency(struct eventtimer *et, uint64_t newfreq)
|
|
|
|
{
|
|
|
|
|
|
|
|
ET_LOCK();
|
|
|
|
if (et == timer) {
|
|
|
|
configtimer(0);
|
|
|
|
et->et_frequency = newfreq;
|
|
|
|
configtimer(1);
|
|
|
|
} else
|
|
|
|
et->et_frequency = newfreq;
|
|
|
|
ET_UNLOCK();
|
|
|
|
}
|
|
|
|
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
void
|
|
|
|
cpu_new_callout(int cpu, sbintime_t bt, sbintime_t bt_opt)
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
{
|
|
|
|
struct pcpu_state *state;
|
|
|
|
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
/* Do not touch anything if somebody reconfiguring timers. */
|
|
|
|
if (busy)
|
|
|
|
return;
|
|
|
|
CTR6(KTR_SPARE2, "new co at %d: on %d at %d.%08x - %d.%08x",
|
|
|
|
curcpu, cpu, (int)(bt_opt >> 32), (u_int)(bt_opt & 0xffffffff),
|
|
|
|
(int)(bt >> 32), (u_int)(bt & 0xffffffff));
|
2017-11-25 23:41:05 +00:00
|
|
|
|
|
|
|
KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
state = DPCPU_ID_PTR(cpu, timerstate);
|
|
|
|
ET_HW_LOCK(state);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
* If there is callout time already set earlier -- do nothing.
|
|
|
|
* This check may appear redundant because we check already in
|
|
|
|
* callout_process() but this double check guarantees we're safe
|
|
|
|
* with respect to race conditions between interrupts execution
|
|
|
|
* and scheduling.
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
*/
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
state->nextcallopt = bt_opt;
|
|
|
|
if (bt >= state->nextcall)
|
|
|
|
goto done;
|
|
|
|
state->nextcall = bt;
|
|
|
|
/* If there is some other event set earlier -- do nothing. */
|
|
|
|
if (bt >= state->nextevent)
|
|
|
|
goto done;
|
|
|
|
state->nextevent = bt;
|
|
|
|
/* If timer is periodic -- there is nothing to reprogram. */
|
|
|
|
if (periodic)
|
|
|
|
goto done;
|
|
|
|
/* If timer is global or of the current CPU -- reprogram it. */
|
|
|
|
if ((timer->et_flags & ET_FLAGS_PERCPU) == 0 || cpu == curcpu) {
|
|
|
|
loadtimer(sbinuptime(), 0);
|
|
|
|
done:
|
|
|
|
ET_HW_UNLOCK(state);
|
|
|
|
return;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
}
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
/* Otherwise make other CPU to reprogram it. */
|
|
|
|
state->handle = 1;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_HW_UNLOCK(state);
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
#ifdef SMP
|
|
|
|
ipi_cpu(cpu, IPI_HARDCLOCK);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
#endif
|
- Make callout(9) tickless, relying on eventtimers(4) as backend for
precise time event generation. This greatly improves granularity of
callouts which are not anymore constrained to wait next tick to be
scheduled.
- Extend the callout KPI introducing a set of callout_reset_sbt* functions,
which take a sbintime_t as timeout argument. The new KPI also offers a
way for consumers to specify precision tolerance they allow, so that
callout can coalesce events and reduce number of interrupts as well as
potentially avoid scheduling a SWI thread.
- Introduce support for dispatching callouts directly from hardware
interrupt context, specifying an additional flag. This feature should be
used carefully, as long as interrupt context has some limitations
(e.g. no sleeping locks can be held).
- Enhance mechanisms to gather informations about callwheel, introducing
a new sysctl to obtain stats.
This change breaks the KBI. struct callout fields has been changed, in
particular 'int ticks' (4 bytes) has been replaced with 'sbintime_t'
(8 bytes) and another 'sbintime_t' field was added for precision.
Together with: mav
Reviewed by: attilio, bde, luigi, phk
Sponsored by: Google Summer of Code 2012, iXsystems inc.
Tested by: flo (amd64, sparc64), marius (sparc64), ian (arm),
markj (amd64), mav, Fabian Keil
2013-03-04 11:09:56 +00:00
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Report or change the active event timers hardware.
|
|
|
|
*/
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
static int
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
sysctl_kern_eventtimer_timer(SYSCTL_HANDLER_ARGS)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
|
|
|
char buf[32];
|
|
|
|
struct eventtimer *et;
|
|
|
|
int error;
|
|
|
|
|
|
|
|
ET_LOCK();
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
et = timer;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
snprintf(buf, sizeof(buf), "%s", et->et_name);
|
|
|
|
ET_UNLOCK();
|
|
|
|
error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
|
|
|
|
ET_LOCK();
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
et = timer;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
if (error != 0 || req->newptr == NULL ||
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
strcasecmp(buf, et->et_name) == 0) {
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
ET_UNLOCK();
|
|
|
|
return (error);
|
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
et = et_find(buf, 0, 0);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
if (et == NULL) {
|
|
|
|
ET_UNLOCK();
|
|
|
|
return (ENOENT);
|
|
|
|
}
|
|
|
|
configtimer(0);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
et_free(timer);
|
|
|
|
if (et->et_flags & ET_FLAGS_C3STOP)
|
2015-01-05 20:44:44 +00:00
|
|
|
cpu_disable_c3_sleep++;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
if (timer->et_flags & ET_FLAGS_C3STOP)
|
2015-01-05 20:44:44 +00:00
|
|
|
cpu_disable_c3_sleep--;
|
2010-09-21 16:50:24 +00:00
|
|
|
periodic = want_periodic;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
timer = et;
|
|
|
|
et_init(timer, timercb, NULL, NULL);
|
|
|
|
configtimer(1);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
ET_UNLOCK();
|
|
|
|
return (error);
|
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
SYSCTL_PROC(_kern_eventtimer, OID_AUTO, timer,
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE,
|
2010-09-14 04:48:04 +00:00
|
|
|
0, 0, sysctl_kern_eventtimer_timer, "A", "Chosen event timer");
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
/*
|
|
|
|
* Report or change the active event timer periodicity.
|
|
|
|
*/
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
static int
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
sysctl_kern_eventtimer_periodic(SYSCTL_HANDLER_ARGS)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
{
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
int error, val;
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
val = periodic;
|
|
|
|
error = sysctl_handle_int(oidp, &val, 0, req);
|
|
|
|
if (error != 0 || req->newptr == NULL)
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
return (error);
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
ET_LOCK();
|
|
|
|
configtimer(0);
|
2010-09-21 16:50:24 +00:00
|
|
|
periodic = want_periodic = val;
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
configtimer(1);
|
Implement new event timers infrastructure. It provides unified APIs for
writing event timer drivers, for choosing best possible drivers by machine
independent code and for operating them to supply kernel with hardclock(),
statclock() and profclock() events in unified fashion on various hardware.
Infrastructure provides support for both per-CPU (independent for every CPU
core) and global timers in periodic and one-shot modes. MI management code
at this moment uses only periodic mode, but one-shot mode use planned for
later, as part of tickless kernel project.
For this moment infrastructure used on i386 and amd64 architectures. Other
archs are welcome to follow, while their current operation should not be
affected.
This patch updates existing drivers (i8254, RTC and LAPIC) for the new
order, and adds event timers support into the HPET driver. These drivers
have different capabilities:
LAPIC - per-CPU timer, supports periodic and one-shot operation, may
freeze in C3 state, calibrated on first use, so may be not exactly precise.
HPET - depending on hardware can work as per-CPU or global, supports
periodic and one-shot operation, usually provides several event timers.
i8254 - global, limited to periodic mode, because same hardware used also
as time counter.
RTC - global, supports only periodic mode, set of frequencies in Hz
limited by powers of 2.
Depending on hardware capabilities, drivers preferred in following orders,
either LAPIC, HPETs, i8254, RTC or HPETs, LAPIC, i8254, RTC.
User may explicitly specify wanted timers via loader tunables or sysctls:
kern.eventtimer.timer1 and kern.eventtimer.timer2.
If requested driver is unavailable or unoperational, system will try to
replace it. If no more timers available or "NONE" specified for second,
system will operate using only one timer, multiplying it's frequency by few
times and uing respective dividers to honor hz, stathz and profhz values,
set during initial setup.
2010-06-20 21:33:29 +00:00
|
|
|
ET_UNLOCK();
|
|
|
|
return (error);
|
|
|
|
}
|
Refactor timer management code with priority to one-shot operation mode.
The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
2010-09-13 07:25:35 +00:00
|
|
|
SYSCTL_PROC(_kern_eventtimer, OID_AUTO, periodic,
|
|
|
|
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
|
2010-09-14 04:48:04 +00:00
|
|
|
0, 0, sysctl_kern_eventtimer_periodic, "I", "Enable event timer periodic mode");
|
2015-02-04 14:49:47 +00:00
|
|
|
|
|
|
|
#include "opt_ddb.h"
|
|
|
|
|
|
|
|
#ifdef DDB
|
|
|
|
#include <ddb/ddb.h>
|
|
|
|
|
|
|
|
DB_SHOW_COMMAND(clocksource, db_show_clocksource)
|
|
|
|
{
|
|
|
|
struct pcpu_state *st;
|
|
|
|
int c;
|
|
|
|
|
|
|
|
CPU_FOREACH(c) {
|
|
|
|
st = DPCPU_ID_PTR(c, timerstate);
|
|
|
|
db_printf(
|
|
|
|
"CPU %2d: action %d handle %d ipi %d idle %d\n"
|
|
|
|
" now %#jx nevent %#jx (%jd)\n"
|
|
|
|
" ntick %#jx (%jd) nhard %#jx (%jd)\n"
|
|
|
|
" nstat %#jx (%jd) nprof %#jx (%jd)\n"
|
|
|
|
" ncall %#jx (%jd) ncallopt %#jx (%jd)\n",
|
|
|
|
c, st->action, st->handle, st->ipi, st->idle,
|
|
|
|
(uintmax_t)st->now,
|
|
|
|
(uintmax_t)st->nextevent,
|
|
|
|
(uintmax_t)(st->nextevent - st->now) / tick_sbt,
|
|
|
|
(uintmax_t)st->nexttick,
|
|
|
|
(uintmax_t)(st->nexttick - st->now) / tick_sbt,
|
|
|
|
(uintmax_t)st->nexthard,
|
|
|
|
(uintmax_t)(st->nexthard - st->now) / tick_sbt,
|
|
|
|
(uintmax_t)st->nextstat,
|
|
|
|
(uintmax_t)(st->nextstat - st->now) / tick_sbt,
|
|
|
|
(uintmax_t)st->nextprof,
|
|
|
|
(uintmax_t)(st->nextprof - st->now) / tick_sbt,
|
|
|
|
(uintmax_t)st->nextcall,
|
|
|
|
(uintmax_t)(st->nextcall - st->now) / tick_sbt,
|
|
|
|
(uintmax_t)st->nextcallopt,
|
|
|
|
(uintmax_t)(st->nextcallopt - st->now) / tick_sbt);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#endif
|