linux/arch/x86/lib/delay.c

140 lines
2.8 KiB
C
Raw Normal View History

/*
* Precise Delay Loops for i386
*
* Copyright (C) 1993 Linus Torvalds
* Copyright (C) 1997 Martin Mares <mj@atrey.karlin.mff.cuni.cz>
* Copyright (C) 2008 Jiri Hladky <hladky _dot_ jiri _at_ gmail _dot_ com>
*
* The __delay function must _NOT_ be inlined as its execution time
* depends wildly on alignment on many x86 processors. The additional
* jump magic is needed to get the timing stable on all the CPU's
* we have to worry about.
*/
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/timex.h>
#include <linux/preempt.h>
#include <linux/delay.h>
#include <asm/processor.h>
#include <asm/delay.h>
#include <asm/timer.h>
#ifdef CONFIG_SMP
# include <asm/smp.h>
#endif
/* simple loop based delay: */
static void delay_loop(unsigned long loops)
{
asm volatile(
" test %0,%0 \n"
" jz 3f \n"
" jmp 1f \n"
".align 16 \n"
"1: jmp 2f \n"
".align 16 \n"
"2: dec %0 \n"
" jnz 2b \n"
"3: dec %0 \n"
: /* we don't need output */
:"a" (loops)
);
}
/* TSC based delay: */
static void delay_tsc(unsigned long __loops)
{
u32 bclock, now, loops = __loops;
x86: enable preemption in delay The RT team has been searching for a nasty latency. This latency shows up out of the blue and has been seen to be as big as 5ms! Using ftrace I found the cause of the latency. pcscd-2995 3dNh1 52360300us : irq_exit (smp_apic_timer_interrupt) pcscd-2995 3dN.2 52360301us : idle_cpu (irq_exit) pcscd-2995 3dN.2 52360301us : rcu_irq_exit (irq_exit) pcscd-2995 3dN.1 52360771us : smp_apic_timer_interrupt (apic_timer_interrupt ) pcscd-2995 3dN.1 52360771us : exit_idle (smp_apic_timer_interrupt) Here's an example of a 400 us latency. pcscd took a timer interrupt and returned with "need resched" enabled, but did not reschedule until after the next interrupt came in at 52360771us 400us later! At first I thought we somehow missed a preemption check in entry.S. But I also noticed that this always seemed to happen during a __delay call. pcscd-2995 3dN.2 52360836us : rcu_irq_exit (irq_exit) pcscd-2995 3.N.. 52361265us : preempt_schedule (__delay) Looking at the x86 delay, I found my problem. In git commit 35d5d08a085c56f153458c3f5d8ce24123617faf, Andrew Morton placed preempt_disable around the entire delay due to TSC's not working nicely on SMP. Unfortunately for those that care about latencies this is devastating! Especially when we have callers to mdelay(8). Here I enable preemption during the loop and account for anytime the task migrates to a new CPU. The delay asked for may be extended a bit by the migration, but delay only guarantees that it will delay for that minimum time. Delaying longer should not be an issue. [ Thanks to Thomas Gleixner for spotting that cpu wasn't updated, and to place the rep_nop between preempt_enabled/disable. ] Signed-off-by: Steven Rostedt <srostedt@redhat.com> Cc: akpm@osdl.org Cc: Clark Williams <clark.williams@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: "Luis Claudio R. Goncalves" <lclaudio@uudg.org> Cc: Gregory Haskins <ghaskins@novell.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andi Kleen <andi-suse@firstfloor.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-25 15:13:32 +00:00
int cpu;
x86: enable preemption in delay The RT team has been searching for a nasty latency. This latency shows up out of the blue and has been seen to be as big as 5ms! Using ftrace I found the cause of the latency. pcscd-2995 3dNh1 52360300us : irq_exit (smp_apic_timer_interrupt) pcscd-2995 3dN.2 52360301us : idle_cpu (irq_exit) pcscd-2995 3dN.2 52360301us : rcu_irq_exit (irq_exit) pcscd-2995 3dN.1 52360771us : smp_apic_timer_interrupt (apic_timer_interrupt ) pcscd-2995 3dN.1 52360771us : exit_idle (smp_apic_timer_interrupt) Here's an example of a 400 us latency. pcscd took a timer interrupt and returned with "need resched" enabled, but did not reschedule until after the next interrupt came in at 52360771us 400us later! At first I thought we somehow missed a preemption check in entry.S. But I also noticed that this always seemed to happen during a __delay call. pcscd-2995 3dN.2 52360836us : rcu_irq_exit (irq_exit) pcscd-2995 3.N.. 52361265us : preempt_schedule (__delay) Looking at the x86 delay, I found my problem. In git commit 35d5d08a085c56f153458c3f5d8ce24123617faf, Andrew Morton placed preempt_disable around the entire delay due to TSC's not working nicely on SMP. Unfortunately for those that care about latencies this is devastating! Especially when we have callers to mdelay(8). Here I enable preemption during the loop and account for anytime the task migrates to a new CPU. The delay asked for may be extended a bit by the migration, but delay only guarantees that it will delay for that minimum time. Delaying longer should not be an issue. [ Thanks to Thomas Gleixner for spotting that cpu wasn't updated, and to place the rep_nop between preempt_enabled/disable. ] Signed-off-by: Steven Rostedt <srostedt@redhat.com> Cc: akpm@osdl.org Cc: Clark Williams <clark.williams@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: "Luis Claudio R. Goncalves" <lclaudio@uudg.org> Cc: Gregory Haskins <ghaskins@novell.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andi Kleen <andi-suse@firstfloor.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-25 15:13:32 +00:00
preempt_disable();
cpu = smp_processor_id();
x86, delay: tsc based udelay should have rdtsc_barrier delay_tsc needs rdtsc_barrier to provide proper delay. Output from a test driver using hpet to cross check delay provided by udelay(). Before: [ 86.794363] Expected delay 5us actual 4679ns [ 87.154362] Expected delay 5us actual 698ns [ 87.514162] Expected delay 5us actual 4539ns [ 88.653716] Expected delay 5us actual 4539ns [ 94.664106] Expected delay 10us actual 9638ns [ 95.049351] Expected delay 10us actual 10126ns [ 95.416110] Expected delay 10us actual 9568ns [ 95.799216] Expected delay 10us actual 9638ns [ 103.624104] Expected delay 10us actual 9707ns [ 104.020619] Expected delay 10us actual 768ns [ 104.419951] Expected delay 10us actual 9707ns After: [ 50.983320] Expected delay 5us actual 5587ns [ 51.261807] Expected delay 5us actual 5587ns [ 51.565715] Expected delay 5us actual 5657ns [ 51.861171] Expected delay 5us actual 5587ns [ 52.164704] Expected delay 5us actual 5726ns [ 52.487457] Expected delay 5us actual 5657ns [ 52.789338] Expected delay 5us actual 5726ns [ 57.119680] Expected delay 10us actual 10755ns [ 57.893997] Expected delay 10us actual 10615ns [ 58.261287] Expected delay 10us actual 10755ns [ 58.620505] Expected delay 10us actual 10825ns [ 58.941035] Expected delay 10us actual 10755ns [ 59.320903] Expected delay 10us actual 10615ns [ 61.306311] Expected delay 10us actual 10755ns [ 61.520542] Expected delay 10us actual 10615ns Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-06-25 23:44:31 +00:00
rdtsc_barrier();
rdtscl(bclock);
x86: enable preemption in delay The RT team has been searching for a nasty latency. This latency shows up out of the blue and has been seen to be as big as 5ms! Using ftrace I found the cause of the latency. pcscd-2995 3dNh1 52360300us : irq_exit (smp_apic_timer_interrupt) pcscd-2995 3dN.2 52360301us : idle_cpu (irq_exit) pcscd-2995 3dN.2 52360301us : rcu_irq_exit (irq_exit) pcscd-2995 3dN.1 52360771us : smp_apic_timer_interrupt (apic_timer_interrupt ) pcscd-2995 3dN.1 52360771us : exit_idle (smp_apic_timer_interrupt) Here's an example of a 400 us latency. pcscd took a timer interrupt and returned with "need resched" enabled, but did not reschedule until after the next interrupt came in at 52360771us 400us later! At first I thought we somehow missed a preemption check in entry.S. But I also noticed that this always seemed to happen during a __delay call. pcscd-2995 3dN.2 52360836us : rcu_irq_exit (irq_exit) pcscd-2995 3.N.. 52361265us : preempt_schedule (__delay) Looking at the x86 delay, I found my problem. In git commit 35d5d08a085c56f153458c3f5d8ce24123617faf, Andrew Morton placed preempt_disable around the entire delay due to TSC's not working nicely on SMP. Unfortunately for those that care about latencies this is devastating! Especially when we have callers to mdelay(8). Here I enable preemption during the loop and account for anytime the task migrates to a new CPU. The delay asked for may be extended a bit by the migration, but delay only guarantees that it will delay for that minimum time. Delaying longer should not be an issue. [ Thanks to Thomas Gleixner for spotting that cpu wasn't updated, and to place the rep_nop between preempt_enabled/disable. ] Signed-off-by: Steven Rostedt <srostedt@redhat.com> Cc: akpm@osdl.org Cc: Clark Williams <clark.williams@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: "Luis Claudio R. Goncalves" <lclaudio@uudg.org> Cc: Gregory Haskins <ghaskins@novell.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andi Kleen <andi-suse@firstfloor.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-25 15:13:32 +00:00
for (;;) {
x86, delay: tsc based udelay should have rdtsc_barrier delay_tsc needs rdtsc_barrier to provide proper delay. Output from a test driver using hpet to cross check delay provided by udelay(). Before: [ 86.794363] Expected delay 5us actual 4679ns [ 87.154362] Expected delay 5us actual 698ns [ 87.514162] Expected delay 5us actual 4539ns [ 88.653716] Expected delay 5us actual 4539ns [ 94.664106] Expected delay 10us actual 9638ns [ 95.049351] Expected delay 10us actual 10126ns [ 95.416110] Expected delay 10us actual 9568ns [ 95.799216] Expected delay 10us actual 9638ns [ 103.624104] Expected delay 10us actual 9707ns [ 104.020619] Expected delay 10us actual 768ns [ 104.419951] Expected delay 10us actual 9707ns After: [ 50.983320] Expected delay 5us actual 5587ns [ 51.261807] Expected delay 5us actual 5587ns [ 51.565715] Expected delay 5us actual 5657ns [ 51.861171] Expected delay 5us actual 5587ns [ 52.164704] Expected delay 5us actual 5726ns [ 52.487457] Expected delay 5us actual 5657ns [ 52.789338] Expected delay 5us actual 5726ns [ 57.119680] Expected delay 10us actual 10755ns [ 57.893997] Expected delay 10us actual 10615ns [ 58.261287] Expected delay 10us actual 10755ns [ 58.620505] Expected delay 10us actual 10825ns [ 58.941035] Expected delay 10us actual 10755ns [ 59.320903] Expected delay 10us actual 10615ns [ 61.306311] Expected delay 10us actual 10755ns [ 61.520542] Expected delay 10us actual 10615ns Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-06-25 23:44:31 +00:00
rdtsc_barrier();
rdtscl(now);
x86: enable preemption in delay The RT team has been searching for a nasty latency. This latency shows up out of the blue and has been seen to be as big as 5ms! Using ftrace I found the cause of the latency. pcscd-2995 3dNh1 52360300us : irq_exit (smp_apic_timer_interrupt) pcscd-2995 3dN.2 52360301us : idle_cpu (irq_exit) pcscd-2995 3dN.2 52360301us : rcu_irq_exit (irq_exit) pcscd-2995 3dN.1 52360771us : smp_apic_timer_interrupt (apic_timer_interrupt ) pcscd-2995 3dN.1 52360771us : exit_idle (smp_apic_timer_interrupt) Here's an example of a 400 us latency. pcscd took a timer interrupt and returned with "need resched" enabled, but did not reschedule until after the next interrupt came in at 52360771us 400us later! At first I thought we somehow missed a preemption check in entry.S. But I also noticed that this always seemed to happen during a __delay call. pcscd-2995 3dN.2 52360836us : rcu_irq_exit (irq_exit) pcscd-2995 3.N.. 52361265us : preempt_schedule (__delay) Looking at the x86 delay, I found my problem. In git commit 35d5d08a085c56f153458c3f5d8ce24123617faf, Andrew Morton placed preempt_disable around the entire delay due to TSC's not working nicely on SMP. Unfortunately for those that care about latencies this is devastating! Especially when we have callers to mdelay(8). Here I enable preemption during the loop and account for anytime the task migrates to a new CPU. The delay asked for may be extended a bit by the migration, but delay only guarantees that it will delay for that minimum time. Delaying longer should not be an issue. [ Thanks to Thomas Gleixner for spotting that cpu wasn't updated, and to place the rep_nop between preempt_enabled/disable. ] Signed-off-by: Steven Rostedt <srostedt@redhat.com> Cc: akpm@osdl.org Cc: Clark Williams <clark.williams@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: "Luis Claudio R. Goncalves" <lclaudio@uudg.org> Cc: Gregory Haskins <ghaskins@novell.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andi Kleen <andi-suse@firstfloor.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-25 15:13:32 +00:00
if ((now - bclock) >= loops)
break;
/* Allow RT tasks to run */
preempt_enable();
rep_nop();
preempt_disable();
/*
* It is possible that we moved to another CPU, and
* since TSC's are per-cpu we need to calculate
* that. The delay must guarantee that we wait "at
* least" the amount of time. Being moved to another
* CPU could make the wait longer but we just need to
* make sure we waited long enough. Rebalance the
* counter for this CPU.
*/
if (unlikely(cpu != smp_processor_id())) {
loops -= (now - bclock);
cpu = smp_processor_id();
x86, delay: tsc based udelay should have rdtsc_barrier delay_tsc needs rdtsc_barrier to provide proper delay. Output from a test driver using hpet to cross check delay provided by udelay(). Before: [ 86.794363] Expected delay 5us actual 4679ns [ 87.154362] Expected delay 5us actual 698ns [ 87.514162] Expected delay 5us actual 4539ns [ 88.653716] Expected delay 5us actual 4539ns [ 94.664106] Expected delay 10us actual 9638ns [ 95.049351] Expected delay 10us actual 10126ns [ 95.416110] Expected delay 10us actual 9568ns [ 95.799216] Expected delay 10us actual 9638ns [ 103.624104] Expected delay 10us actual 9707ns [ 104.020619] Expected delay 10us actual 768ns [ 104.419951] Expected delay 10us actual 9707ns After: [ 50.983320] Expected delay 5us actual 5587ns [ 51.261807] Expected delay 5us actual 5587ns [ 51.565715] Expected delay 5us actual 5657ns [ 51.861171] Expected delay 5us actual 5587ns [ 52.164704] Expected delay 5us actual 5726ns [ 52.487457] Expected delay 5us actual 5657ns [ 52.789338] Expected delay 5us actual 5726ns [ 57.119680] Expected delay 10us actual 10755ns [ 57.893997] Expected delay 10us actual 10615ns [ 58.261287] Expected delay 10us actual 10755ns [ 58.620505] Expected delay 10us actual 10825ns [ 58.941035] Expected delay 10us actual 10755ns [ 59.320903] Expected delay 10us actual 10615ns [ 61.306311] Expected delay 10us actual 10755ns [ 61.520542] Expected delay 10us actual 10615ns Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-06-25 23:44:31 +00:00
rdtsc_barrier();
x86: enable preemption in delay The RT team has been searching for a nasty latency. This latency shows up out of the blue and has been seen to be as big as 5ms! Using ftrace I found the cause of the latency. pcscd-2995 3dNh1 52360300us : irq_exit (smp_apic_timer_interrupt) pcscd-2995 3dN.2 52360301us : idle_cpu (irq_exit) pcscd-2995 3dN.2 52360301us : rcu_irq_exit (irq_exit) pcscd-2995 3dN.1 52360771us : smp_apic_timer_interrupt (apic_timer_interrupt ) pcscd-2995 3dN.1 52360771us : exit_idle (smp_apic_timer_interrupt) Here's an example of a 400 us latency. pcscd took a timer interrupt and returned with "need resched" enabled, but did not reschedule until after the next interrupt came in at 52360771us 400us later! At first I thought we somehow missed a preemption check in entry.S. But I also noticed that this always seemed to happen during a __delay call. pcscd-2995 3dN.2 52360836us : rcu_irq_exit (irq_exit) pcscd-2995 3.N.. 52361265us : preempt_schedule (__delay) Looking at the x86 delay, I found my problem. In git commit 35d5d08a085c56f153458c3f5d8ce24123617faf, Andrew Morton placed preempt_disable around the entire delay due to TSC's not working nicely on SMP. Unfortunately for those that care about latencies this is devastating! Especially when we have callers to mdelay(8). Here I enable preemption during the loop and account for anytime the task migrates to a new CPU. The delay asked for may be extended a bit by the migration, but delay only guarantees that it will delay for that minimum time. Delaying longer should not be an issue. [ Thanks to Thomas Gleixner for spotting that cpu wasn't updated, and to place the rep_nop between preempt_enabled/disable. ] Signed-off-by: Steven Rostedt <srostedt@redhat.com> Cc: akpm@osdl.org Cc: Clark Williams <clark.williams@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: "Luis Claudio R. Goncalves" <lclaudio@uudg.org> Cc: Gregory Haskins <ghaskins@novell.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andi Kleen <andi-suse@firstfloor.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-25 15:13:32 +00:00
rdtscl(bclock);
}
}
preempt_enable();
}
/*
* Since we calibrate only once at boot, this
* function should be set once at boot and not changed
*/
static void (*delay_fn)(unsigned long) = delay_loop;
void use_tsc_delay(void)
{
delay_fn = delay_tsc;
}
int read_current_timer(unsigned long *timer_val)
{
if (delay_fn == delay_tsc) {
rdtscll(*timer_val);
return 0;
}
return -1;
}
void __delay(unsigned long loops)
{
delay_fn(loops);
}
EXPORT_SYMBOL(__delay);
inline void __const_udelay(unsigned long xloops)
{
int d0;
xloops *= 4;
asm("mull %%edx"
:"=d" (xloops), "=&a" (d0)
:"1" (xloops), "0"
(this_cpu_read(cpu_info.loops_per_jiffy) * (HZ/4)));
__delay(++xloops);
}
EXPORT_SYMBOL(__const_udelay);
void __udelay(unsigned long usecs)
{
__const_udelay(usecs * 0x000010c7); /* 2**32 / 1000000 (rounded up) */
}
EXPORT_SYMBOL(__udelay);
void __ndelay(unsigned long nsecs)
{
__const_udelay(nsecs * 0x00005); /* 2**32 / 1000000000 (rounded up) */
}
EXPORT_SYMBOL(__ndelay);