Skip to contentSkip to conversation

I doublt on timer interrput.

question liyu 3 participants 6 messages Nov 7, 2005 – Nov 8, 2005

lore
Share & tools
lore

liyu

author
#0 lore raw

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~

#1 lore raw
Quoted 11 lines · liyu

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():

Quoted 10 lines

++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

author
#2 lore raw

Fawad Lateef Wrote:

Quoted 45 lines · liyu
Quoted 12 lines

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():

Quoted 12 lines

++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 !

Quoted 11 lines

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

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

#3 lore raw
Quoted 42 lines · liyu

Fawad Lateef Wrote:

Quoted 11 lines

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.

Quoted 9 lines

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);
Quoted 8 lines

#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

Quoted 7 lines

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

Quoted 8 lines

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) :)

Quoted 4 lines

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

Quoted 8 lines

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,

--
Fawad Lateef

author
#4 lore raw

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:

Quoted 16 lines

[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

Quoted 121 lines · Fawad Lateef
Quoted 45 lines · liyu

Fawad Lateef Wrote:

Quoted 10 lines

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.

Quoted 11 lines

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);
Quoted 10 lines

#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

Quoted 7 lines

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

Quoted 10 lines

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) :)

Quoted 6 lines

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

Quoted 9 lines

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,

--
Fawad Lateef

#5 lore raw
In reply to liyu
Quoted 12 lines · 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():

Quoted 8 lines

++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