[question] I doublt on timer interrput.

[liyu]

Hi, all:

I have one question about timer interrupt (i386 architecture).

As we known, the timer emit HZ times interrputs per second,
and in i386. The interrupt handler will call scheduler_tick()
each time (on i386 at least, both enable or disable APIC).

On my Celeron machine(IOW, only one CPU, not SMP/SMT), I defined
a global int variable 'tick_count' in kernel/sched.c, and add one line
of code like follow in scheduler_tick():

++tick_count;

but I found it is not same with content of the /proc/interrupts,
and the differennt between them is not little.

I can not understand why that is.

Any useful idea.

-liyu / NOW~

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majo...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

[Fawad Lateef]

On 11/7/05, liyu <li...@ccoss.com.cn> wrote:
>
> I have one question about timer interrupt (i386 architecture).
>
> As we known, the timer emit HZ times interrputs per second,
> and in i386. The interrupt handler will call scheduler_tick()
> each time (on i386 at least, both enable or disable APIC).
>
> On my Celeron machine(IOW, only one CPU, not SMP/SMT), I defined
> a global int variable 'tick_count' in kernel/sched.c, and add one line
> of code like follow in scheduler_tick():
>
> ++tick_count;
>
> but I found it is not same with content of the /proc/interrupts,
> and the differennt between them is not little.
>
> I can not understand why that is.
>
> Any useful idea.
>
>

What I found in the kernel code is that scheduler_tick is called from
two locations in the kernel (2.6.14-mm1) code (i386).

1) from kernel/timer.c in update_process_times which is called from
arch/i386/kernel/apic.c and its calling depends on the CONFIG_SMP
defined or not (see
http://sosdg.org/~coywolf/lxr/source/arch/i386/kernel/apic.c#L1160)
and as you don't have CONFIG_SMP enabled so its won't be called from
here.

2) from sched_fork function in kernel/sched.c
(http://sosdg.org/~coywolf/lxr/source/kernel/sched.c#L1414) and I
think its called when newly forked process setup is going to be
performed, and I think as from here scheduler_tick is called in your
case, so you are getting different value for your variable tick_count

scheduler_tick might be called from somewhere else which I am missing
so please CMIIW !

--
Fawad Lateef

[Robert Love]

On Mon, 2005-11-07 at 14:05 +0800, liyu wrote:
> Hi, all:
>
> I have one question about timer interrupt (i386 architecture).
>
> As we known, the timer emit HZ times interrputs per second,
> and in i386. The interrupt handler will call scheduler_tick()
> each time (on i386 at least, both enable or disable APIC).
>
> On my Celeron machine(IOW, only one CPU, not SMP/SMT), I defined
> a global int variable 'tick_count' in kernel/sched.c, and add one line
> of code like follow in scheduler_tick():
>
> ++tick_count;
>
> but I found it is not same with content of the /proc/interrupts,
> and the differennt between them is not little.
>
> I can not understand why that is.
>
> Any useful idea.

scheduler_tick() is not the timer interrupt.

You want to hook into do_timer() or similar.

Robert Love

[liyu]

Fawad Lateef Wrote:

Please see this URL:

http://lxr.linux.no/source/include/asm-i386/mach-default/do_timer.h#L20

static inline void do_timer_interrupt_hook(struct pt_regs *regs)
{
do_timer(regs);
#ifndef CONFIG_SMP
update_process_times(user_mode(regs));
#endif
/*
* In the SMP case we use the local APIC timer interrupt to do the
* profiling, except when we simulate SMP mode on a uniprocessor
* system, in that case we have to call the local interrupt handler.
*/
#ifndef CONFIG_X86_LOCAL_APIC
profile_tick(CPU_PROFILING, regs);
#else
if (!using_apic_timer)
smp_local_timer_interrupt(regs);
#endif
}

That is the code in 2.6.12, but 2.6.13.3 also same with it at least.
So we call scheduler_tick() HZ times per second, both enable
SMP or disable it.

Nod, I agree with your words, the scheduler_tick() do not same with
timer interrupt handler on call times. but I guess it should be more
than jiffies, beacause of other functions also can call it (for example,
as Lateef said, sched_fork().)

I think that

scheduler_tick() might be called from somewhere

is not exact.

We may note, it do not be EXPORT_SYMBOL_*()ed , so it only can be called
from kernel core,
not kernel modules. Such a few places we can find it use LXR or grep.

I use setup one sysctl integer variable to watch the value of 'count_tick',
Do this way have any problem? I found some value skips, but I think it is
normal case.

However, I will make a experiemnt that write one hook like do_timer(),
as Love said

PS: if our scheduler_tick() is not called every timer interrput, the
compute of task timeslice
also is not exact ?!

Thanks a lot.

-liyu

[Fawad Lateef]

On 11/8/05, liyu <li...@ccoss.com.cn> wrote:

> Fawad Lateef Wrote:
>
> >
> >What I found in the kernel code is that scheduler_tick is called from
> >two locations in the kernel (2.6.14-mm1) code (i386).
> >
> >1) from kernel/timer.c in update_process_times which is called from
> >arch/i386/kernel/apic.c and its calling depends on the CONFIG_SMP
> >defined or not (see
> >http://sosdg.org/~coywolf/lxr/source/arch/i386/kernel/apic.c#L1160)
> >and as you don't have CONFIG_SMP enabled so its won't be called from
> >here.
> >
> >2) from sched_fork function in kernel/sched.c
> >(http://sosdg.org/~coywolf/lxr/source/kernel/sched.c#L1414) and I
> >think its called when newly forked process setup is going to be
> >performed, and I think as from here scheduler_tick is called in your
> >case, so you are getting different value for your variable tick_count
> >
> >scheduler_tick might be called from somewhere else which I am missing
> >so please CMIIW !
> >

> Please see this URL:
>
> http://lxr.linux.no/source/include/asm-i386/mach-default/do_timer.h#L20
>
> static inline void do_timer_interrupt_hook(struct pt_regs *regs)
> {
> do_timer(regs);
> #ifndef CONFIG_SMP
> update_process_times(user_mode(regs));
> #endif
> /*
> * In the SMP case we use the local APIC timer interrupt to do the
> * profiling, except when we simulate SMP mode on a uniprocessor
> * system, in that case we have to call the local interrupt handler.
> */
> #ifndef CONFIG_X86_LOCAL_APIC
> profile_tick(CPU_PROFILING, regs);
> #else
> if (!using_apic_timer)
> smp_local_timer_interrupt(regs);
> #endif
> }
>
>
> That is the code in 2.6.12, but 2.6.13.3 also same with it at least.
> So we call scheduler_tick() HZ times per second, both enable
> SMP or disable it.
>

Yes, this is the thing which I missed

> Nod, I agree with your words, the scheduler_tick() do not same with
> timer interrupt handler on call times. but I guess it should be more
> than jiffies, beacause of other functions also can call it (for example,
> as Lateef said, sched_fork().)
>
> I think that
>
> scheduler_tick() might be called from somewhere
>
> is not exact.
>
> We may note, it do not be EXPORT_SYMBOL_*()ed , so it only can be called
> from kernel core,
> not kernel modules. Such a few places we can find it use LXR or grep.
>

By saying __might_be_called_from_somewhere__ I meant that I am missing
some-other place __with-in_the_kernel_code__ from where it is called,
which you pointed to me (about do_timer.h) :)

> I use setup one sysctl integer variable to watch the value of 'count_tick',
> Do this way have any problem? I found some value skips, but I think it is
> normal case.
>

If you are declaring count_tick as a global variable (without static)
in sched.c then you can just use it in your test module by specifying
extern for your variable

>
> However, I will make a experiemnt that write one hook like do_timer(),
> as Love said
>
> PS: if our scheduler_tick() is not called every timer interrput, the
> compute of task timeslice
> also is not exact ?!
>

Yes, I am now sure that it will be called for every timer interrupt ! :)

Thanks,

[liyu]

I get it !

I am sorry to spend your time so much, all trouble are came from one my
low-level fault.

I put "++count_tick" in scheduler_tick(), but that function() can return
before it!

This is one sample result:

> [root@CCOSS_629884359 root]# cat /proc/sys/kernel/count_tick
> /proc/interrupts
> 157491
> CPU0
> 0: 157390 IO-APIC-edge timer
> 1: 10 IO-APIC-edge i8042
> 8: 1 IO-APIC-edge rtc
> 9: 0 IO-APIC-level acpi
> 12: 111 IO-APIC-edge i8042
> 14: 57616 IO-APIC-edge ide0
> 15: 2 IO-APIC-edge ide1
> 16: 0 IO-APIC-level uhci_hcd:usb1
> 17: 0 IO-APIC-level uhci_hcd:usb2
> 18: 893 IO-APIC-level eth0
> NMI: 0
> LOC: 157326
> ERR: 0

We can see:

157491 > 157390 !

Yeah, I got it.

Thanks a lot again.

-liyu