linux/arch/powerpc/kernel/time.c

1220 lines
32 KiB
C
Raw Normal View History

/*
* Common time routines among all ppc machines.
*
* Written by Cort Dougan (cort@cs.nmt.edu) to merge
* Paul Mackerras' version and mine for PReP and Pmac.
* MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
* Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
*
* First round of bugfixes by Gabriel Paubert (paubert@iram.es)
* to make clock more stable (2.4.0-test5). The only thing
* that this code assumes is that the timebases have been synchronized
* by firmware on SMP and are never stopped (never do sleep
* on SMP then, nap and doze are OK).
*
* Speeded up do_gettimeofday by getting rid of references to
* xtime (which required locks for consistency). (mikejc@us.ibm.com)
*
* TODO (not necessarily in this file):
* - improve precision and reproducibility of timebase frequency
* measurement at boot time. (for iSeries, we calibrate the timebase
* against the Titan chip's clock.)
* - for astronomical applications: add a new function to get
* non ambiguous timestamps even around leap seconds. This needs
* a new timestamp format and a good name.
*
* 1997-09-10 Updated NTP code according to technical memorandum Jan '96
* "A Kernel Model for Precision Timekeeping" by Dave Mills
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/errno.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/param.h>
#include <linux/string.h>
#include <linux/mm.h>
#include <linux/interrupt.h>
#include <linux/timex.h>
#include <linux/kernel_stat.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/profile.h>
#include <linux/cpu.h>
#include <linux/security.h>
#include <linux/percpu.h>
#include <linux/rtc.h>
#include <linux/jiffies.h>
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
#include <linux/posix-timers.h>
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 13:55:46 +00:00
#include <linux/irq.h>
#include <linux/delay.h>
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
#include <linux/perf_event.h>
#include <asm/io.h>
#include <asm/processor.h>
#include <asm/nvram.h>
#include <asm/cache.h>
#include <asm/machdep.h>
#include <asm/uaccess.h>
#include <asm/time.h>
#include <asm/prom.h>
#include <asm/irq.h>
#include <asm/div64.h>
#include <asm/smp.h>
#include <asm/vdso_datapage.h>
#include <asm/firmware.h>
#include <asm/cputime.h>
#ifdef CONFIG_PPC_ISERIES
#include <asm/iseries/it_lp_queue.h>
#include <asm/iseries/hv_call_xm.h>
#endif
/* powerpc clocksource/clockevent code */
#include <linux/clockchips.h>
#include <linux/clocksource.h>
static cycle_t rtc_read(struct clocksource *);
static struct clocksource clocksource_rtc = {
.name = "rtc",
.rating = 400,
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
.mask = CLOCKSOURCE_MASK(64),
.shift = 22,
.mult = 0, /* To be filled in */
.read = rtc_read,
};
static cycle_t timebase_read(struct clocksource *);
static struct clocksource clocksource_timebase = {
.name = "timebase",
.rating = 400,
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
.mask = CLOCKSOURCE_MASK(64),
.shift = 22,
.mult = 0, /* To be filled in */
.read = timebase_read,
};
#define DECREMENTER_MAX 0x7fffffff
static int decrementer_set_next_event(unsigned long evt,
struct clock_event_device *dev);
static void decrementer_set_mode(enum clock_event_mode mode,
struct clock_event_device *dev);
static struct clock_event_device decrementer_clockevent = {
.name = "decrementer",
.rating = 200,
.shift = 0, /* To be filled in */
.mult = 0, /* To be filled in */
.irq = 0,
.set_next_event = decrementer_set_next_event,
.set_mode = decrementer_set_mode,
.features = CLOCK_EVT_FEAT_ONESHOT,
};
struct decrementer_clock {
struct clock_event_device event;
u64 next_tb;
};
static DEFINE_PER_CPU(struct decrementer_clock, decrementers);
#ifdef CONFIG_PPC_ISERIES
static unsigned long __initdata iSeries_recal_titan;
static signed long __initdata iSeries_recal_tb;
/* Forward declaration is only needed for iSereis compiles */
static void __init clocksource_init(void);
#endif
#define XSEC_PER_SEC (1024*1024)
#ifdef CONFIG_PPC64
#define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
#else
/* compute ((xsec << 12) * max) >> 32 */
#define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
#endif
unsigned long tb_ticks_per_jiffy;
unsigned long tb_ticks_per_usec = 100; /* sane default */
EXPORT_SYMBOL(tb_ticks_per_usec);
unsigned long tb_ticks_per_sec;
EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
u64 tb_to_xs;
unsigned tb_to_us;
#define TICKLEN_SCALE NTP_SCALE_SHIFT
static u64 last_tick_len; /* units are ns / 2^TICKLEN_SCALE */
static u64 ticklen_to_xs; /* 0.64 fraction */
/* If last_tick_len corresponds to about 1/HZ seconds, then
last_tick_len << TICKLEN_SHIFT will be about 2^63. */
#define TICKLEN_SHIFT (63 - 30 - TICKLEN_SCALE + SHIFT_HZ)
DEFINE_SPINLOCK(rtc_lock);
EXPORT_SYMBOL_GPL(rtc_lock);
static u64 tb_to_ns_scale __read_mostly;
static unsigned tb_to_ns_shift __read_mostly;
static unsigned long boot_tb __read_mostly;
extern struct timezone sys_tz;
static long timezone_offset;
unsigned long ppc_proc_freq;
EXPORT_SYMBOL(ppc_proc_freq);
unsigned long ppc_tb_freq;
static u64 tb_last_jiffy __cacheline_aligned_in_smp;
static DEFINE_PER_CPU(u64, last_jiffy);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
#ifdef CONFIG_VIRT_CPU_ACCOUNTING
/*
* Factors for converting from cputime_t (timebase ticks) to
* jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds).
* These are all stored as 0.64 fixed-point binary fractions.
*/
u64 __cputime_jiffies_factor;
EXPORT_SYMBOL(__cputime_jiffies_factor);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
u64 __cputime_msec_factor;
EXPORT_SYMBOL(__cputime_msec_factor);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
u64 __cputime_sec_factor;
EXPORT_SYMBOL(__cputime_sec_factor);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
u64 __cputime_clockt_factor;
EXPORT_SYMBOL(__cputime_clockt_factor);
DEFINE_PER_CPU(unsigned long, cputime_last_delta);
DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
cputime_t cputime_one_jiffy;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
static void calc_cputime_factors(void)
{
struct div_result res;
div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
__cputime_jiffies_factor = res.result_low;
div128_by_32(1000, 0, tb_ticks_per_sec, &res);
__cputime_msec_factor = res.result_low;
div128_by_32(1, 0, tb_ticks_per_sec, &res);
__cputime_sec_factor = res.result_low;
div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
__cputime_clockt_factor = res.result_low;
}
/*
* Read the PURR on systems that have it, otherwise the timebase.
*/
static u64 read_purr(void)
{
if (cpu_has_feature(CPU_FTR_PURR))
return mfspr(SPRN_PURR);
return mftb();
}
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
/*
* Read the SPURR on systems that have it, otherwise the purr
*/
static u64 read_spurr(u64 purr)
{
/*
* cpus without PURR won't have a SPURR
* We already know the former when we use this, so tell gcc
*/
if (cpu_has_feature(CPU_FTR_PURR) && cpu_has_feature(CPU_FTR_SPURR))
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
return mfspr(SPRN_SPURR);
return purr;
}
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
/*
* Account time for a transition between system, hard irq
* or soft irq state.
*/
void account_system_vtime(struct task_struct *tsk)
{
u64 now, nowscaled, delta, deltascaled, sys_time;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
unsigned long flags;
local_irq_save(flags);
now = read_purr();
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
nowscaled = read_spurr(now);
delta = now - get_paca()->startpurr;
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
deltascaled = nowscaled - get_paca()->startspurr;
get_paca()->startpurr = now;
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
get_paca()->startspurr = nowscaled;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
if (!in_interrupt()) {
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
/* deltascaled includes both user and system time.
* Hence scale it based on the purr ratio to estimate
* the system time */
sys_time = get_paca()->system_time;
if (get_paca()->user_time)
deltascaled = deltascaled * sys_time /
(sys_time + get_paca()->user_time);
delta += sys_time;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
get_paca()->system_time = 0;
}
2008-12-31 14:11:38 +00:00
if (in_irq() || idle_task(smp_processor_id()) != tsk)
account_system_time(tsk, 0, delta, deltascaled);
else
account_idle_time(delta);
per_cpu(cputime_last_delta, smp_processor_id()) = delta;
per_cpu(cputime_scaled_last_delta, smp_processor_id()) = deltascaled;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
local_irq_restore(flags);
}
/*
* Transfer the user and system times accumulated in the paca
* by the exception entry and exit code to the generic process
* user and system time records.
* Must be called with interrupts disabled.
*/
void account_process_tick(struct task_struct *tsk, int user_tick)
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
{
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
cputime_t utime, utimescaled;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
utime = get_paca()->user_time;
get_paca()->user_time = 0;
utimescaled = cputime_to_scaled(utime);
account_user_time(tsk, utime, utimescaled);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
}
/*
* Stuff for accounting stolen time.
*/
struct cpu_purr_data {
int initialized; /* thread is running */
u64 tb; /* last TB value read */
u64 purr; /* last PURR value read */
powerpc: add scaled time accounting This adds POWERPC specific hooks for scaled time accounting. POWER6 includes a SPURR register. The SPURR is based off the PURR register but is scaled based on CPU frequency and issue rates. This gives a more accurate account of the instructions used per task. The PURR and timebase will be constant relative to the wall clock, irrespective of the CPU frequency. This implementation reads the SPURR register in account_system_vtime which is only call called on context witch and hard and soft irq entry and exit. The percentage of user and system time is then estimated using the ratio of these accounted by the PURR. If the SPURR is not present, the PURR read. An earlier implementation of this patch read the SPURR whenever the PURR was read, which included the system call entry and exit path. Unfortunately this showed a performance regression on lmbench runs, so was re-implemented. I've included the lmbench results here when run bare metal on POWER6. 1st column is the unpatch results. 2nd column is the results using the below patch and the 3rd is the % diff of these results from the base. 4th and 5th columns are the results and % differnce from the base using the older patch (SPURR read in syscall entry/exit path). Base Scaled-Acct SPURR-in-syscall Result Result % diff Result % diff Simple syscall: 0.3086 0.3086 0.0000 0.3452 11.8600 Simple read: 0.4591 0.4671 1.7425 0.5044 9.86713 Simple write: 0.4364 0.4366 0.0458 0.4731 8.40971 Simple stat: 2.0055 2.0295 1.1967 2.0669 3.06158 Simple fstat: 0.5962 0.5876 -1.442 0.6368 6.80979 Simple open/close: 3.1283 3.1009 -0.875 3.2088 2.57328 Select on 10 fd's: 0.8554 0.8457 -1.133 0.8667 1.32101 Select on 100 fd's: 3.5292 3.6329 2.9383 3.6664 3.88756 Select on 250 fd's: 7.9097 8.1881 3.5197 8.2242 3.97613 Select on 500 fd's: 15.2659 15.836 3.7357 15.873 3.97814 Select on 10 tcp fd's: 0.9576 0.9416 -1.670 0.9752 1.83792 Select on 100 tcp fd's: 7.248 7.2254 -0.311 7.2685 0.28283 Select on 250 tcp fd's: 17.7742 17.707 -0.375 17.749 -0.1406 Select on 500 tcp fd's: 35.4258 35.25 -0.496 35.286 -0.3929 Signal handler installation: 0.6131 0.6075 -0.913 0.647 5.52927 Signal handler overhead: 2.0919 2.1078 0.7600 2.1831 4.35967 Protection fault: 0.7345 0.7478 1.8107 0.8031 9.33968 Pipe latency: 33.006 16.398 -50.31 33.475 1.42368 AF_UNIX sock stream latency: 14.5093 30.910 113.03 30.715 111.692 Process fork+exit: 219.8 222.8 1.3648 229.37 4.35623 Process fork+execve: 876.14 873.28 -0.32 868.66 -0.8533 Process fork+/bin/sh -c: 2830 2876.5 1.6431 2958 4.52296 File /var/tmp/XXX write bw: 1193497 1195536 0.1708 118657 -0.5799 Pagefaults on /var/tmp/XXX: 3.1272 3.2117 2.7020 3.2521 3.99398 Also, kernel compile times show no difference with this patch applied. [pbadari@us.ibm.com: Avoid unnecessary PURR reading] Signed-off-by: Michael Neuling <mikey@neuling.org> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Jay Lan <jlan@engr.sgi.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-18 10:06:37 +00:00
u64 spurr; /* last SPURR value read */
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
};
/*
* Each entry in the cpu_purr_data array is manipulated only by its
* "owner" cpu -- usually in the timer interrupt but also occasionally
* in process context for cpu online. As long as cpus do not touch
* each others' cpu_purr_data, disabling local interrupts is
* sufficient to serialize accesses.
*/
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data);
static void snapshot_tb_and_purr(void *data)
{
unsigned long flags;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data);
local_irq_save(flags);
p->tb = get_tb_or_rtc();
p->purr = mfspr(SPRN_PURR);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
wmb();
p->initialized = 1;
local_irq_restore(flags);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
}
/*
* Called during boot when all cpus have come up.
*/
void snapshot_timebases(void)
{
if (!cpu_has_feature(CPU_FTR_PURR))
return;
on_each_cpu(snapshot_tb_and_purr, NULL, 1);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
}
/*
* Must be called with interrupts disabled.
*/
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
void calculate_steal_time(void)
{
u64 tb, purr;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
s64 stolen;
struct cpu_purr_data *pme;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
pme = &__get_cpu_var(cpu_purr_data);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
if (!pme->initialized)
return; /* !CPU_FTR_PURR or early in early boot */
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
tb = mftb();
purr = mfspr(SPRN_PURR);
stolen = (tb - pme->tb) - (purr - pme->purr);
2008-12-31 14:11:38 +00:00
if (stolen > 0) {
if (idle_task(smp_processor_id()) != current)
account_steal_time(stolen);
else
account_idle_time(stolen);
}
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
pme->tb = tb;
pme->purr = purr;
}
#ifdef CONFIG_PPC_SPLPAR
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
/*
* Must be called before the cpu is added to the online map when
* a cpu is being brought up at runtime.
*/
static void snapshot_purr(void)
{
struct cpu_purr_data *pme;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
unsigned long flags;
if (!cpu_has_feature(CPU_FTR_PURR))
return;
local_irq_save(flags);
pme = &__get_cpu_var(cpu_purr_data);
pme->tb = mftb();
pme->purr = mfspr(SPRN_PURR);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
pme->initialized = 1;
local_irq_restore(flags);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
}
#endif /* CONFIG_PPC_SPLPAR */
#else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
#define calc_cputime_factors()
#define calculate_steal_time() do { } while (0)
#endif
#if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR))
#define snapshot_purr() do { } while (0)
#endif
/*
* Called when a cpu comes up after the system has finished booting,
* i.e. as a result of a hotplug cpu action.
*/
void snapshot_timebase(void)
{
__get_cpu_var(last_jiffy) = get_tb_or_rtc();
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
snapshot_purr();
}
void __delay(unsigned long loops)
{
unsigned long start;
int diff;
if (__USE_RTC()) {
start = get_rtcl();
do {
/* the RTCL register wraps at 1000000000 */
diff = get_rtcl() - start;
if (diff < 0)
diff += 1000000000;
} while (diff < loops);
} else {
start = get_tbl();
while (get_tbl() - start < loops)
HMT_low();
HMT_medium();
}
}
EXPORT_SYMBOL(__delay);
void udelay(unsigned long usecs)
{
__delay(tb_ticks_per_usec * usecs);
}
EXPORT_SYMBOL(udelay);
static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec,
u64 new_tb_to_xs)
{
/*
* tb_update_count is used to allow the userspace gettimeofday code
* to assure itself that it sees a consistent view of the tb_to_xs and
* stamp_xsec variables. It reads the tb_update_count, then reads
* tb_to_xs and stamp_xsec and then reads tb_update_count again. If
* the two values of tb_update_count match and are even then the
* tb_to_xs and stamp_xsec values are consistent. If not, then it
* loops back and reads them again until this criteria is met.
* We expect the caller to have done the first increment of
* vdso_data->tb_update_count already.
*/
vdso_data->tb_orig_stamp = new_tb_stamp;
vdso_data->stamp_xsec = new_stamp_xsec;
vdso_data->tb_to_xs = new_tb_to_xs;
vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec;
vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec;
vdso_data->stamp_xtime = xtime;
smp_wmb();
++(vdso_data->tb_update_count);
}
#ifdef CONFIG_SMP
unsigned long profile_pc(struct pt_regs *regs)
{
unsigned long pc = instruction_pointer(regs);
if (in_lock_functions(pc))
return regs->link;
return pc;
}
EXPORT_SYMBOL(profile_pc);
#endif
#ifdef CONFIG_PPC_ISERIES
/*
* This function recalibrates the timebase based on the 49-bit time-of-day
* value in the Titan chip. The Titan is much more accurate than the value
* returned by the service processor for the timebase frequency.
*/
static int __init iSeries_tb_recal(void)
{
struct div_result divres;
unsigned long titan, tb;
/* Make sure we only run on iSeries */
if (!firmware_has_feature(FW_FEATURE_ISERIES))
return -ENODEV;
tb = get_tb();
titan = HvCallXm_loadTod();
if ( iSeries_recal_titan ) {
unsigned long tb_ticks = tb - iSeries_recal_tb;
unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
unsigned long new_tb_ticks_per_sec = (tb_ticks * USEC_PER_SEC)/titan_usec;
unsigned long new_tb_ticks_per_jiffy =
DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
char sign = '+';
/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
if ( tick_diff < 0 ) {
tick_diff = -tick_diff;
sign = '-';
}
if ( tick_diff ) {
if ( tick_diff < tb_ticks_per_jiffy/25 ) {
printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
new_tb_ticks_per_jiffy, sign, tick_diff );
tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
tb_ticks_per_sec = new_tb_ticks_per_sec;
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
calc_cputime_factors();
div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres );
tb_to_xs = divres.result_low;
vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
vdso_data->tb_to_xs = tb_to_xs;
setup_cputime_one_jiffy();
}
else {
printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
" new tb_ticks_per_jiffy = %lu\n"
" old tb_ticks_per_jiffy = %lu\n",
new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
}
}
}
iSeries_recal_titan = titan;
iSeries_recal_tb = tb;
/* Called here as now we know accurate values for the timebase */
clocksource_init();
return 0;
}
late_initcall(iSeries_tb_recal);
/* Called from platform early init */
void __init iSeries_time_init_early(void)
{
iSeries_recal_tb = get_tb();
iSeries_recal_titan = HvCallXm_loadTod();
}
#endif /* CONFIG_PPC_ISERIES */
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_PPC32)
DEFINE_PER_CPU(u8, perf_event_pending);
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
void set_perf_event_pending(void)
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
{
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
get_cpu_var(perf_event_pending) = 1;
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
set_dec(1);
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
put_cpu_var(perf_event_pending);
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
}
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
#define test_perf_event_pending() __get_cpu_var(perf_event_pending)
#define clear_perf_event_pending() __get_cpu_var(perf_event_pending) = 0
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
#else /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
#define test_perf_event_pending() 0
#define clear_perf_event_pending()
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
#endif /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
/*
* For iSeries shared processors, we have to let the hypervisor
* set the hardware decrementer. We set a virtual decrementer
* in the lppaca and call the hypervisor if the virtual
* decrementer is less than the current value in the hardware
* decrementer. (almost always the new decrementer value will
* be greater than the current hardware decementer so the hypervisor
* call will not be needed)
*/
/*
* timer_interrupt - gets called when the decrementer overflows,
* with interrupts disabled.
*/
void timer_interrupt(struct pt_regs * regs)
{
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 13:55:46 +00:00
struct pt_regs *old_regs;
struct decrementer_clock *decrementer = &__get_cpu_var(decrementers);
struct clock_event_device *evt = &decrementer->event;
u64 now;
/* Ensure a positive value is written to the decrementer, or else
* some CPUs will continuue to take decrementer exceptions */
set_dec(DECREMENTER_MAX);
#ifdef CONFIG_PPC32
perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\<event\>/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian <eranian@google.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Kyle McMartin <kyle@mcmartin.ca> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> LKML-Reference: <new-submission> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 10:02:48 +00:00
if (test_perf_event_pending()) {
clear_perf_event_pending();
perf_event_do_pending();
perf_counter: powerpc: Enable use of software counters on 32-bit powerpc This enables the perf_counter subsystem on 32-bit powerpc. Since we don't have any support for hardware counters on 32-bit powerpc yet, only software counters can be used. Besides selecting HAVE_PERF_COUNTERS for 32-bit powerpc as well as 64-bit, the main thing this does is add an implementation of set_perf_counter_pending(). This needs to arrange for perf_counter_do_pending() to be called when interrupts are enabled. Rather than add code to local_irq_restore as 64-bit does, the 32-bit set_perf_counter_pending() generates an interrupt by setting the decrementer to 1 so that a decrementer interrupt will become pending in 1 or 2 timebase ticks (if a decrementer interrupt isn't already pending). When interrupts are enabled, timer_interrupt() will be called, and some new code in there calls perf_counter_do_pending(). We use a per-cpu array of flags to indicate whether we need to call perf_counter_do_pending() or not. This introduces a couple of new Kconfig symbols: PPC_HAVE_PMU_SUPPORT, which is selected by processor families for which we have hardware PMU support (currently only PPC64), and PPC_PERF_CTRS, which enables the powerpc-specific perf_counter back-end. Signed-off-by: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: linuxppc-dev@ozlabs.org Cc: benh@kernel.crashing.org LKML-Reference: <19000.55404.103840.393470@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-06-17 11:50:04 +00:00
}
if (atomic_read(&ppc_n_lost_interrupts) != 0)
do_IRQ(regs);
#endif
now = get_tb_or_rtc();
if (now < decrementer->next_tb) {
/* not time for this event yet */
now = decrementer->next_tb - now;
if (now <= DECREMENTER_MAX)
set_dec((int)now);
return;
}
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 13:55:46 +00:00
old_regs = set_irq_regs(regs);
irq_enter();
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
calculate_steal_time();
#ifdef CONFIG_PPC_ISERIES
if (firmware_has_feature(FW_FEATURE_ISERIES))
get_lppaca()->int_dword.fields.decr_int = 0;
#endif
if (evt->event_handler)
evt->event_handler(evt);
#ifdef CONFIG_PPC_ISERIES
if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
process_hvlpevents();
#endif
#ifdef CONFIG_PPC64
/* collect purr register values often, for accurate calculations */
if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
cu->current_tb = mfspr(SPRN_PURR);
}
#endif
irq_exit();
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 13:55:46 +00:00
set_irq_regs(old_regs);
}
void wakeup_decrementer(void)
{
unsigned long ticks;
/*
* The timebase gets saved on sleep and restored on wakeup,
* so all we need to do is to reset the decrementer.
*/
ticks = tb_ticks_since(__get_cpu_var(last_jiffy));
if (ticks < tb_ticks_per_jiffy)
ticks = tb_ticks_per_jiffy - ticks;
else
ticks = 1;
set_dec(ticks);
}
#ifdef CONFIG_SUSPEND
void generic_suspend_disable_irqs(void)
{
preempt_disable();
/* Disable the decrementer, so that it doesn't interfere
* with suspending.
*/
set_dec(0x7fffffff);
local_irq_disable();
set_dec(0x7fffffff);
}
void generic_suspend_enable_irqs(void)
{
wakeup_decrementer();
local_irq_enable();
preempt_enable();
}
/* Overrides the weak version in kernel/power/main.c */
void arch_suspend_disable_irqs(void)
{
if (ppc_md.suspend_disable_irqs)
ppc_md.suspend_disable_irqs();
generic_suspend_disable_irqs();
}
/* Overrides the weak version in kernel/power/main.c */
void arch_suspend_enable_irqs(void)
{
generic_suspend_enable_irqs();
if (ppc_md.suspend_enable_irqs)
ppc_md.suspend_enable_irqs();
}
#endif
#ifdef CONFIG_SMP
void __init smp_space_timers(unsigned int max_cpus)
{
int i;
u64 previous_tb = per_cpu(last_jiffy, boot_cpuid);
/* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
previous_tb -= tb_ticks_per_jiffy;
for_each_possible_cpu(i) {
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
if (i == boot_cpuid)
continue;
per_cpu(last_jiffy, i) = previous_tb;
}
}
#endif
/*
* Scheduler clock - returns current time in nanosec units.
*
* Note: mulhdu(a, b) (multiply high double unsigned) returns
* the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
* are 64-bit unsigned numbers.
*/
unsigned long long sched_clock(void)
{
if (__USE_RTC())
return get_rtc();
return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
}
static int __init get_freq(char *name, int cells, unsigned long *val)
{
struct device_node *cpu;
const unsigned int *fp;
int found = 0;
/* The cpu node should have timebase and clock frequency properties */
cpu = of_find_node_by_type(NULL, "cpu");
if (cpu) {
fp = of_get_property(cpu, name, NULL);
if (fp) {
found = 1;
*val = of_read_ulong(fp, cells);
}
of_node_put(cpu);
}
return found;
}
/* should become __cpuinit when secondary_cpu_time_init also is */
void start_cpu_decrementer(void)
{
#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
/* Clear any pending timer interrupts */
mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
/* Enable decrementer interrupt */
mtspr(SPRN_TCR, TCR_DIE);
#endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
}
void __init generic_calibrate_decr(void)
{
ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
!get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
printk(KERN_ERR "WARNING: Estimating decrementer frequency "
"(not found)\n");
}
ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
!get_freq("clock-frequency", 1, &ppc_proc_freq)) {
printk(KERN_ERR "WARNING: Estimating processor frequency "
"(not found)\n");
}
}
int update_persistent_clock(struct timespec now)
{
struct rtc_time tm;
if (!ppc_md.set_rtc_time)
return 0;
to_tm(now.tv_sec + 1 + timezone_offset, &tm);
tm.tm_year -= 1900;
tm.tm_mon -= 1;
return ppc_md.set_rtc_time(&tm);
}
void read_persistent_clock(struct timespec *ts)
{
struct rtc_time tm;
static int first = 1;
ts->tv_nsec = 0;
/* XXX this is a litle fragile but will work okay in the short term */
if (first) {
first = 0;
if (ppc_md.time_init)
timezone_offset = ppc_md.time_init();
/* get_boot_time() isn't guaranteed to be safe to call late */
if (ppc_md.get_boot_time) {
ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
return;
}
}
if (!ppc_md.get_rtc_time) {
ts->tv_sec = 0;
return;
}
ppc_md.get_rtc_time(&tm);
ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
tm.tm_hour, tm.tm_min, tm.tm_sec);
}
/* clocksource code */
static cycle_t rtc_read(struct clocksource *cs)
{
return (cycle_t)get_rtc();
}
static cycle_t timebase_read(struct clocksource *cs)
{
return (cycle_t)get_tb();
}
void update_vsyscall(struct timespec *wall_time, struct clocksource *clock)
{
u64 t2x, stamp_xsec;
if (clock != &clocksource_timebase)
return;
/* Make userspace gettimeofday spin until we're done. */
++vdso_data->tb_update_count;
smp_mb();
/* XXX this assumes clock->shift == 22 */
/* 4611686018 ~= 2^(20+64-22) / 1e9 */
t2x = (u64) clock->mult * 4611686018ULL;
stamp_xsec = (u64) xtime.tv_nsec * XSEC_PER_SEC;
do_div(stamp_xsec, 1000000000);
stamp_xsec += (u64) xtime.tv_sec * XSEC_PER_SEC;
update_gtod(clock->cycle_last, stamp_xsec, t2x);
}
void update_vsyscall_tz(void)
{
/* Make userspace gettimeofday spin until we're done. */
++vdso_data->tb_update_count;
smp_mb();
vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
vdso_data->tz_dsttime = sys_tz.tz_dsttime;
smp_mb();
++vdso_data->tb_update_count;
}
static void __init clocksource_init(void)
{
struct clocksource *clock;
if (__USE_RTC())
clock = &clocksource_rtc;
else
clock = &clocksource_timebase;
clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
if (clocksource_register(clock)) {
printk(KERN_ERR "clocksource: %s is already registered\n",
clock->name);
return;
}
printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
clock->name, clock->mult, clock->shift);
}
static int decrementer_set_next_event(unsigned long evt,
struct clock_event_device *dev)
{
__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
set_dec(evt);
return 0;
}
static void decrementer_set_mode(enum clock_event_mode mode,
struct clock_event_device *dev)
{
if (mode != CLOCK_EVT_MODE_ONESHOT)
decrementer_set_next_event(DECREMENTER_MAX, dev);
}
static void __init setup_clockevent_multiplier(unsigned long hz)
{
u64 mult, shift = 32;
while (1) {
mult = div_sc(hz, NSEC_PER_SEC, shift);
if (mult && (mult >> 32UL) == 0UL)
break;
shift--;
}
decrementer_clockevent.shift = shift;
decrementer_clockevent.mult = mult;
}
static void register_decrementer_clockevent(int cpu)
{
struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
*dec = decrementer_clockevent;
dec->cpumask = cpumask_of(cpu);
printk(KERN_DEBUG "clockevent: %s mult[%lx] shift[%d] cpu[%d]\n",
dec->name, dec->mult, dec->shift, cpu);
clockevents_register_device(dec);
}
static void __init init_decrementer_clockevent(void)
{
int cpu = smp_processor_id();
setup_clockevent_multiplier(ppc_tb_freq);
decrementer_clockevent.max_delta_ns =
clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
decrementer_clockevent.min_delta_ns =
clockevent_delta2ns(2, &decrementer_clockevent);
register_decrementer_clockevent(cpu);
}
void secondary_cpu_time_init(void)
{
/* Start the decrementer on CPUs that have manual control
* such as BookE
*/
start_cpu_decrementer();
/* FIME: Should make unrelatred change to move snapshot_timebase
* call here ! */
register_decrementer_clockevent(smp_processor_id());
}
/* This function is only called on the boot processor */
void __init time_init(void)
{
unsigned long flags;
struct div_result res;
u64 scale, x;
unsigned shift;
if (__USE_RTC()) {
/* 601 processor: dec counts down by 128 every 128ns */
ppc_tb_freq = 1000000000;
tb_last_jiffy = get_rtcl();
} else {
/* Normal PowerPC with timebase register */
ppc_md.calibrate_decr();
printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
tb_last_jiffy = get_tb();
}
tb_ticks_per_jiffy = ppc_tb_freq / HZ;
tb_ticks_per_sec = ppc_tb_freq;
tb_ticks_per_usec = ppc_tb_freq / 1000000;
tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000);
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-23 23:06:59 +00:00
calc_cputime_factors();
setup_cputime_one_jiffy();
/*
* Calculate the length of each tick in ns. It will not be
* exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ.
* We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq,
* rounded up.
*/
x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1;
do_div(x, ppc_tb_freq);
tick_nsec = x;
last_tick_len = x << TICKLEN_SCALE;
/*
* Compute ticklen_to_xs, which is a factor which gets multiplied
* by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value.
* It is computed as:
* ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9)
* where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT
* which turns out to be N = 51 - SHIFT_HZ.
* This gives the result as a 0.64 fixed-point fraction.
* That value is reduced by an offset amounting to 1 xsec per
* 2^31 timebase ticks to avoid problems with time going backwards
* by 1 xsec when we do timer_recalc_offset due to losing the
* fractional xsec. That offset is equal to ppc_tb_freq/2^51
* since there are 2^20 xsec in a second.
*/
div128_by_32((1ULL << 51) - ppc_tb_freq, 0,
tb_ticks_per_jiffy << SHIFT_HZ, &res);
div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res);
ticklen_to_xs = res.result_low;
/* Compute tb_to_xs from tick_nsec */
tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs);
/*
* Compute scale factor for sched_clock.
* The calibrate_decr() function has set tb_ticks_per_sec,
* which is the timebase frequency.
* We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
* the 128-bit result as a 64.64 fixed-point number.
* We then shift that number right until it is less than 1.0,
* giving us the scale factor and shift count to use in
* sched_clock().
*/
div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
scale = res.result_low;
for (shift = 0; res.result_high != 0; ++shift) {
scale = (scale >> 1) | (res.result_high << 63);
res.result_high >>= 1;
}
tb_to_ns_scale = scale;
tb_to_ns_shift = shift;
/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
boot_tb = get_tb_or_rtc();
write_seqlock_irqsave(&xtime_lock, flags);
/* If platform provided a timezone (pmac), we correct the time */
if (timezone_offset) {
sys_tz.tz_minuteswest = -timezone_offset / 60;
sys_tz.tz_dsttime = 0;
}
vdso_data->tb_orig_stamp = tb_last_jiffy;
vdso_data->tb_update_count = 0;
vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
vdso_data->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC;
vdso_data->tb_to_xs = tb_to_xs;
write_sequnlock_irqrestore(&xtime_lock, flags);
/* Start the decrementer on CPUs that have manual control
* such as BookE
*/
start_cpu_decrementer();
/* Register the clocksource, if we're not running on iSeries */
if (!firmware_has_feature(FW_FEATURE_ISERIES))
clocksource_init();
init_decrementer_clockevent();
}
#define FEBRUARY 2
#define STARTOFTIME 1970
#define SECDAY 86400L
#define SECYR (SECDAY * 365)
#define leapyear(year) ((year) % 4 == 0 && \
((year) % 100 != 0 || (year) % 400 == 0))
#define days_in_year(a) (leapyear(a) ? 366 : 365)
#define days_in_month(a) (month_days[(a) - 1])
static int month_days[12] = {
31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
};
/*
* This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
*/
void GregorianDay(struct rtc_time * tm)
{
int leapsToDate;
int lastYear;
int day;
int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
lastYear = tm->tm_year - 1;
/*
* Number of leap corrections to apply up to end of last year
*/
leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
/*
* This year is a leap year if it is divisible by 4 except when it is
* divisible by 100 unless it is divisible by 400
*
* e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
*/
day = tm->tm_mon > 2 && leapyear(tm->tm_year);
day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
tm->tm_mday;
tm->tm_wday = day % 7;
}
void to_tm(int tim, struct rtc_time * tm)
{
register int i;
register long hms, day;
day = tim / SECDAY;
hms = tim % SECDAY;
/* Hours, minutes, seconds are easy */
tm->tm_hour = hms / 3600;
tm->tm_min = (hms % 3600) / 60;
tm->tm_sec = (hms % 3600) % 60;
/* Number of years in days */
for (i = STARTOFTIME; day >= days_in_year(i); i++)
day -= days_in_year(i);
tm->tm_year = i;
/* Number of months in days left */
if (leapyear(tm->tm_year))
days_in_month(FEBRUARY) = 29;
for (i = 1; day >= days_in_month(i); i++)
day -= days_in_month(i);
days_in_month(FEBRUARY) = 28;
tm->tm_mon = i;
/* Days are what is left over (+1) from all that. */
tm->tm_mday = day + 1;
/*
* Determine the day of week
*/
GregorianDay(tm);
}
/* Auxiliary function to compute scaling factors */
/* Actually the choice of a timebase running at 1/4 the of the bus
* frequency giving resolution of a few tens of nanoseconds is quite nice.
* It makes this computation very precise (27-28 bits typically) which
* is optimistic considering the stability of most processor clock
* oscillators and the precision with which the timebase frequency
* is measured but does not harm.
*/
unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale)
{
unsigned mlt=0, tmp, err;
/* No concern for performance, it's done once: use a stupid
* but safe and compact method to find the multiplier.
*/
for (tmp = 1U<<31; tmp != 0; tmp >>= 1) {
if (mulhwu(inscale, mlt|tmp) < outscale)
mlt |= tmp;
}
/* We might still be off by 1 for the best approximation.
* A side effect of this is that if outscale is too large
* the returned value will be zero.
* Many corner cases have been checked and seem to work,
* some might have been forgotten in the test however.
*/
err = inscale * (mlt+1);
if (err <= inscale/2)
mlt++;
return mlt;
}
/*
* Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
* result.
*/
void div128_by_32(u64 dividend_high, u64 dividend_low,
unsigned divisor, struct div_result *dr)
{
unsigned long a, b, c, d;
unsigned long w, x, y, z;
u64 ra, rb, rc;
a = dividend_high >> 32;
b = dividend_high & 0xffffffff;
c = dividend_low >> 32;
d = dividend_low & 0xffffffff;
w = a / divisor;
ra = ((u64)(a - (w * divisor)) << 32) + b;
rb = ((u64) do_div(ra, divisor) << 32) + c;
x = ra;
rc = ((u64) do_div(rb, divisor) << 32) + d;
y = rb;
do_div(rc, divisor);
z = rc;
dr->result_high = ((u64)w << 32) + x;
dr->result_low = ((u64)y << 32) + z;
}
/* We don't need to calibrate delay, we use the CPU timebase for that */
void calibrate_delay(void)
{
/* Some generic code (such as spinlock debug) use loops_per_jiffy
* as the number of __delay(1) in a jiffy, so make it so
*/
loops_per_jiffy = tb_ticks_per_jiffy;
}
static int __init rtc_init(void)
{
struct platform_device *pdev;
if (!ppc_md.get_rtc_time)
return -ENODEV;
pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
if (IS_ERR(pdev))
return PTR_ERR(pdev);
return 0;
}
module_init(rtc_init);