LinuxLists.cc - PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Tue, 10 Jun 2008 21:52:23 +0200
solsTiCe d'Hiver <[email protected]> wrote:

> hi,
>
> i am running a 2.6.25.5-rt6 kernel.
> i know there is no such thing as 2.6.25.5-rt6 yet. so i have checked
> with a 2.6.25.4-rt6 kernel and i have the same issue.
>
> when i try to run jackd (0.109.2) i get that strange error:
> FATAL: cannot locate cpu MHz in /proc/cpuinfo
>
> and it appears that /proc/cpuinfo lack the line
> cpu MHz:

btw this is scary; the Mhz line in /proc/cpuinfo doesn't have much
meaning really, so applications using it and depending on it is a
rather big disaster ;-(

Does anyone know why jackd wants to use this?

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-10 20:43:44

by Kok, Auke

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

Arjan van de Ven wrote:
> On Tue, 10 Jun 2008 21:52:23 +0200
> solsTiCe d'Hiver <[email protected]> wrote:
>
>> hi,
>>
>> i am running a 2.6.25.5-rt6 kernel.
>> i know there is no such thing as 2.6.25.5-rt6 yet. so i have checked
>> with a 2.6.25.4-rt6 kernel and i have the same issue.
>>
>> when i try to run jackd (0.109.2) i get that strange error:
>> FATAL: cannot locate cpu MHz in /proc/cpuinfo
>>
>> and it appears that /proc/cpuinfo lack the line
>> cpu MHz:
>
>
> btw this is scary; the Mhz line in /proc/cpuinfo doesn't have much
> meaning really, so applications using it and depending on it is a
> rather big disaster ;-(
>
> Does anyone know why jackd wants to use this?

I just poked into Jackd and it seems to be using fallback code to calculate timing
intervals from the cpu MHz value (config/os/gnu-linux/time.c).

the code itself is self-addmittedly a kludge:

122 /*
123 * This is another kludge. It looks CPU-dependent, but actually it
124 * reflects the lack of standards for the Linux kernel formatting of
125 * /proc/cpuinfo.
126 */
127
128 jack_time_t
129 jack_get_mhz (void)
130 {
131 FILE *f = fopen("/proc/cpuinfo", "r");

one kludge building on another :)

Auke

2008-06-11 12:40:45

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Tuesday 2008-06-10 21:52, solsTiCe d'Hiver wrote:
>hi,
>
>i am running a 2.6.25.5-rt6 kernel.
>i know there is no such thing as 2.6.25.5-rt6 yet. so i have checked with a
>2.6.25.4-rt6 kernel and i have the same issue.

I have 2.6.25.6-rt6 here, so:

processor : 0
vendor_id : AuthenticAMD
cpu family : 6
model : 8
model name : AMD Athlon(tm) XP 2000+
stepping : 0
cpu MHz : 1666.737

it's there.

2008-06-11 12:42:50

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Tuesday 2008-06-10 22:42, Kok, Auke wrote:
>>>
>>> and it appears that /proc/cpuinfo lack the line
>>> cpu MHz:
>>
>>
>> btw this is scary; the Mhz line in /proc/cpuinfo doesn't have much
>> meaning really, so applications using it and depending on it is a
>> rather big disaster ;-(
>>
>> Does anyone know why jackd wants to use this?

It would be nice to have a /sys file where the CPU frequency is in,
but unless you have cpufreq loaded, no such file exists.
And there are quite a few processors which do not have cpufreq,
so /proc/cpuinfo is the only information (besides dmesg) where
you can read it from.

2008-06-11 13:36:52

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 14:42:40 +0200 (CEST)
Jan Engelhardt <[email protected]> wrote:

>
> On Tuesday 2008-06-10 22:42, Kok, Auke wrote:
> >>>
> >>> and it appears that /proc/cpuinfo lack the line
> >>> cpu MHz:
> >>
> >>
> >> btw this is scary; the Mhz line in /proc/cpuinfo doesn't have much
> >> meaning really, so applications using it and depending on it is a
> >> rather big disaster ;-(
> >>
> >> Does anyone know why jackd wants to use this?
>
> It would be nice to have a /sys file where the CPU frequency is in,

may I ask why? What would you do with it?
If it says 100Mhz, what would you conclude? That you have a slow cpu?
Or that you have a really fast CPU that currently runs at a low
frequency because you're mostly idle and are saving power?

CPU frequency is...
1) not something that is fixed; it changes all the time
2) not a good indication for performance; remember the 3.4Ghz pentium4,
a 2Ghz Core2Duo outperforms it
3) something which the cpu itself barely honors (the clock stops during
idle etc etc)

Userland apps that use "cpu frequency" seem to be buggy to me, at
least I don't believe that they understand the complexities about what
this means, and I don't believe that the decisions they make based on
this will be made correctly.

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-11 13:53:14

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wednesday 2008-06-11 15:36, Arjan van de Ven wrote:
>> >> btw this is scary; the Mhz line in /proc/cpuinfo doesn't have much
>> >> meaning really, so applications using it and depending on it is a
>> >> rather big disaster ;-(
>> >>
>> >> Does anyone know why jackd wants to use this?
>>
>> It would be nice to have a /sys file where the CPU frequency is in,
>
>may I ask why? What would you do with it?

Off the top off the head, just for reporting to the user. Desklets and stuff.

>If it says 100Mhz, what would you conclude? That you have a slow cpu?
>Or that you have a really fast CPU that currently runs at a low
>frequency because you're mostly idle and are saving power?
>
>CPU frequency is...
>1) not something that is fixed; it changes all the time
>2) not a good indication for performance; remember the 3.4Ghz pentium4,
>a 2Ghz Core2Duo outperforms it
>3) something which the cpu itself barely honors (the clock stops during
>idle etc etc)

2008-06-11 13:58:30

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 15:52:57 +0200 (CEST)
Jan Engelhardt <[email protected]> wrote:

>
> On Wednesday 2008-06-11 15:36, Arjan van de Ven wrote:
> >> >> btw this is scary; the Mhz line in /proc/cpuinfo doesn't have
> >> >> much meaning really, so applications using it and depending on
> >> >> it is a rather big disaster ;-(
> >> >>
> >> >> Does anyone know why jackd wants to use this?
> >>
> >> It would be nice to have a /sys file where the CPU frequency is in,
> >
> >may I ask why? What would you do with it?
>
> Off the top off the head, just for reporting to the user. Desklets
> and stuff.

even then it doesn't mean much ;-(

> >CPU frequency is...
> >1) not something that is fixed; it changes all the time
> >2) not a good indication for performance; remember the 3.4Ghz
> >pentium4, a 2Ghz Core2Duo outperforms it
> >3) something which the cpu itself barely honors (the clock stops
> >during idle etc etc)

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-11 14:15:28

by solsTiCe d'Hiver

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

> I have 2.6.25.6-rt6 here, so:

> cpu MHz : 1666.737
>
> it's there.

if i remember well it was there too with 2.6.25.4-rt5. i mean with _rt5_
patchset.
but i am not sure and can't test it now because rt5 patch is not available
anymore...

2008-06-11 14:16:35

by linux-os (Dick Johnson)

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008, Arjan van de Ven wrote:

> On Wed, 11 Jun 2008 15:52:57 +0200 (CEST)
> Jan Engelhardt <[email protected]> wrote:
>
>>
>> On Wednesday 2008-06-11 15:36, Arjan van de Ven wrote:
>>>>>> btw this is scary; the Mhz line in /proc/cpuinfo doesn't have
>>>>>> much meaning really, so applications using it and depending on
>>>>>> it is a rather big disaster ;-(
>>>>>>
>>>>>> Does anyone know why jackd wants to use this?
>>>>
>>>> It would be nice to have a /sys file where the CPU frequency is in,
>>>
>>> may I ask why? What would you do with it?
>>
>> Off the top off the head, just for reporting to the user. Desklets
>> and stuff.
>
> even then it doesn't mean much ;-(
>
>>> CPU frequency is...
>>> 1) not something that is fixed; it changes all the time
>>> 2) not a good indication for performance; remember the 3.4Ghz
>>> pentium4, a 2Ghz Core2Duo outperforms it
>>> 3) something which the cpu itself barely honors (the clock stops
>>> during idle etc etc)
>
>

I use CPU frequency to scale the output of rdtsc when timing
routines being optimized. If CPU frequency goes away, I'll
have to add calibration routines to a lot of code. I don't
think one should just dispose of something that has been
useful for many years because they don't understand its
utility. The fact that __some__ machines dynamically adjust
clock-speeds is not a valid reason to remove something useful
that impacts only boot time. Following this trend, soon there
won't be any floppy-drive code because "nobody" uses them
anymore. Please don't keep throwing good stuff away.

Cheers,
Dick Johnson
Penguin : Linux version 2.6.22.1 on an i686 machine (5588.29 BogoMips).
My book : http://www.AbominableFirebug.com/
_

****************************************************************
The information transmitted in this message is confidential and may be privileged. Any review, retransmission, dissemination, or other use of this information by persons or entities other than the intended recipient is prohibited. If you are not the intended recipient, please notify Analogic Corporation immediately - by replying to this message or by sending an email to [email protected] - and destroy all copies of this information, including any attachments, without reading or disclosing them.

Thank you.

2008-06-11 14:37:00

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wednesday 2008-06-11 16:15, solsTiCe d'Hiver wrote:

>> I have 2.6.25.6-rt6 here, so:
>
>> cpu MHz : 1666.737
>>
>> it's there.
>
>if i remember well it was there too with 2.6.25.4-rt5. i mean with _rt5_
>patchset.
>but i am not sure and can't test it now because rt5 patch is not available
>anymore...
>

says who? http://rt.et.redhat.com/download/older/patch-2.6.25.4-rt5.bz2

2008-06-11 15:11:19

by Peter Zijlstra

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 2008-06-11 at 10:14 -0400, linux-os (Dick Johnson) wrote:
> Following this trend, soon there
> won't be any floppy-drive code because "nobody" uses them
> anymore. Please don't keep throwing good stuff away.

floppy and good in a single breath,.. wow

2008-06-11 15:35:35

by solsTiCe d'Hiver

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

> says who? http://rt.et.redhat.com/download/older/patch-2.6.25.4-rt5.bz2
oops.

so i have tested 2.6.25.4-rt {3,4,5}
and all the 3 have the same issue (no cpu MHz in /proc/cpuinfo here)

my hardware is a dell inspiron 531 (with bios 1.0.9) with Athlon 64 x2 4400+
as you already know, a motherboard with MCP61 chipset.

2008-06-11 15:37:07

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 10:14:50 -0400
"linux-os (Dick Johnson)" <[email protected]> wrote:

>
> I use CPU frequency to scale the output of rdtsc when timing
> routines being optimized.

Ok this is broken in various really bad ways.

1) CPU frequency is very fluctuating, it changes in principle every
millisecond or more
2) the rdtsc "frequency" is conceptually unrelated to cpu frequency. In
fact, you'll be hard-pressed to buy a system today where this
relationship works....

> If CPU frequency goes away, I'll
> have to add calibration routines to a lot of code. I don't
> think one should just dispose of something that has been
> useful for many years because they don't understand its
> utility.

clearly you don't understand it either ;-)

I'm not arguing for removing it, at all. I'm just arguing that any user
of it (being human or program) has some problems in how they use the
data.

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-11 16:08:48

by Ray Lee

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, Jun 11, 2008 at 6:36 AM, Arjan van de Ven <[email protected]> wrote:
> CPU frequency is...
> 1) not something that is fixed; it changes all the time

Not necessarily true of setups where jackd is running.

> 2) not a good indication for performance; remember the 3.4Ghz pentium4,
> a 2Ghz Core2Duo outperforms it

Your point only holds if jackd is somehow using this to compare
performance across different cpus. It's not.

> 3) something which the cpu itself barely honors (the clock stops during
> idle etc etc)

Again, that won't be happening when jackd is running.

> Userland apps that use "cpu frequency" seem to be buggy to me, at
> least I don't believe that they understand the complexities about what
> this means,

It doesn't appear you understand what jackd is and what lengths users
will go to to configure their systems for running it smoothly and
reliably.

> and I don't believe that the decisions they make based on
> this will be made correctly.

Please go read the jackd designs slides from the 2003 presentation on
the jack website [1]. The author has a clue, and is quite aware of how
computers work.

[1] http://lad.linuxaudio.org/events/2003_zkm/slides/paul_davis-jack/title.html

2008-06-11 16:32:55

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 09:08:40 -0700
"Ray Lee" <[email protected]> wrote:

> On Wed, Jun 11, 2008 at 6:36 AM, Arjan van de Ven
> <[email protected]> wrote:
> > CPU frequency is...
> > 1) not something that is fixed; it changes all the time
>
> Not necessarily true of setups where jackd is running.

in case you didn't realize, your cpu itself will change it's frequency
under you, even if you don't use things like ondemand cpufreq governor.
(Intel calls this "Intel Dynamic Acceleration technology", and it's
already in all current Intel based laptops)

> Again, that won't be happening when jackd is running.
>
> > Userland apps that use "cpu frequency" seem to be buggy to me, at
> > least I don't believe that they understand the complexities about
> > what this means,
>
> It doesn't appear you understand what jackd is and what lengths users
> will go to to configure their systems for running it smoothly and
> reliably.

I understand that jackd tries to squeeze the last bit out of things.
But really, the cpu frequency doesn't fit into that because it's
dynamic.

>
> > and I don't believe that the decisions they make based on
> > this will be made correctly.
>
> Please go read the jackd designs slides from the 2003 presentation on
> the jack website [1]. The author has a clue, and is quite aware of how
> computers work

I read this now; I don't see anything about using the CPU frequency in
there.....

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-11 17:00:54

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

Arjan van de Ven wrote:
> On Wed, 11 Jun 2008 10:14:50 -0400
> "linux-os (Dick Johnson)" <[email protected]> wrote:
>
>> I use CPU frequency to scale the output of rdtsc when timing
>> routines being optimized.
>
> Ok this is broken in various really bad ways.
>
> 1) CPU frequency is very fluctuating, it changes in principle every
> millisecond or more

I'm curious. What do you mean by "in principle"? And why will it change?

> 2) the rdtsc "frequency" is conceptually unrelated to cpu frequency. In
> fact, you'll be hard-pressed to buy a system today where this
> relationship works....
>

And what do you mean by "conceptually unrelated to cpu frequency"?
Is it not the clock freq that is driving the cpu and the freq at which
the tsc is incremented?

>
>> If CPU frequency goes away, I'll
>> have to add calibration routines to a lot of code. I don't
>> think one should just dispose of something that has been
>> useful for many years because they don't understand its
>> utility.
>
> clearly you don't understand it either ;-)
>

Could you explain what we don't understand about this please?

> I'm not arguing for removing it, at all. I'm just arguing that any user
> of it (being human or program) has some problems in how they use the
> data.
>
>
>

Regards
Mark

2008-06-11 17:45:21

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 12:59:24 -0400
Mark Hounschell <[email protected]> wrote:

> Arjan van de Ven wrote:
> > On Wed, 11 Jun 2008 10:14:50 -0400
> > "linux-os (Dick Johnson)" <[email protected]> wrote:
> >
> >> I use CPU frequency to scale the output of rdtsc when timing
> >> routines being optimized.
> >
> > Ok this is broken in various really bad ways.
> >
> > 1) CPU frequency is very fluctuating, it changes in principle every
> > millisecond or more
>
> I'm curious. What do you mean by "in principle"? And why will it
> change?

ondemand governor will change the cpu frequency dynamically all the
time.
the cpu itself has a dynamic range in which it operates (at least on
cpus that support Intel Dynamic Acceleration technology, IDA)

>
> > 2) the rdtsc "frequency" is conceptually unrelated to cpu
> > frequency. In fact, you'll be hard-pressed to buy a system today
> > where this relationship works....
> >
>
> And what do you mean by "conceptually unrelated to cpu frequency"?
> Is it not the clock freq that is driving the cpu and the freq at which
> the tsc is incremented?

no it is not that... at all.
the tsc comes from an entirely different clock, and on anything you can
buy today from AMD or Intel (or the last year for that matter), it's
fixed frequency (except in idle) irrespective of the frequency the CPU
is operating at!
it's a "time stamp counter" not a "cpu cycle counter". You can run
instructions faster than the tsc increment or slower. Or sometimes at
the same rate.
Which it is depends on what cpufreq/ondemand are doing and how active
IDA is.

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-11 17:56:05

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

Arjan van de Ven wrote:
> On Wed, 11 Jun 2008 12:59:24 -0400
> Mark Hounschell <[email protected]> wrote:
>
>> Arjan van de Ven wrote:
>>> On Wed, 11 Jun 2008 10:14:50 -0400
>>> "linux-os (Dick Johnson)" <[email protected]> wrote:
>>>
>>>> I use CPU frequency to scale the output of rdtsc when timing
>>>> routines being optimized.
>>> Ok this is broken in various really bad ways.
>>>
>>> 1) CPU frequency is very fluctuating, it changes in principle every
>>> millisecond or more
>> I'm curious. What do you mean by "in principle"? And why will it
>> change?
>
> ondemand governor will change the cpu frequency dynamically all the
> time.
> the cpu itself has a dynamic range in which it operates (at least on
> cpus that support Intel Dynamic Acceleration technology, IDA)
>

OK. If I am always using AMD Opteron/X64 class machines am I safe here
when no cpu freq scaling stuff is on and I'm pinned to a particular processor?

>>> 2) the rdtsc "frequency" is conceptually unrelated to cpu
>>> frequency. In fact, you'll be hard-pressed to buy a system today
>>> where this relationship works....
>>>
>> And what do you mean by "conceptually unrelated to cpu frequency"?
>> Is it not the clock freq that is driving the cpu and the freq at which
>> the tsc is incremented?
>
> no it is not that... at all.
> the tsc comes from an entirely different clock, and on anything you can
> buy today from AMD or Intel (or the last year for that matter), it's
> fixed frequency (except in idle) irrespective of the frequency the CPU
> is operating at!
> it's a "time stamp counter" not a "cpu cycle counter". You can run
> instructions faster than the tsc increment or slower. Or sometimes at
> the same rate.
> Which it is depends on what cpufreq/ondemand are doing and how active
> IDA is.
>

I was under the impression that it was the same clock on AMD Opteron and
newer processors.

So if all this is true how and why can the kernel use it but user land is wrong
for doing so?

Mark

2008-06-11 20:31:11

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 13:54:33 -0400
Mark Hounschell <[email protected]> wrote:

> >
> > ondemand governor will change the cpu frequency dynamically all the
> > time.
> > the cpu itself has a dynamic range in which it operates (at least on
> > cpus that support Intel Dynamic Acceleration technology, IDA)
> >
>
> OK. If I am always using AMD Opteron/X64 class machines am I safe here
> when no cpu freq scaling stuff is on and I'm pinned to a particular
> processor?

today, yes. What AMD will do in the next stepping? Only they will know.

>
> >>> 2) the rdtsc "frequency" is conceptually unrelated to cpu
> >>> frequency. In fact, you'll be hard-pressed to buy a system today
> >>> where this relationship works....
> >>>
> >> And what do you mean by "conceptually unrelated to cpu frequency"?
> >> Is it not the clock freq that is driving the cpu and the freq at
> >> which the tsc is incremented?
> >
> > no it is not that... at all.
> > the tsc comes from an entirely different clock, and on anything you
> > can buy today from AMD or Intel (or the last year for that matter),
> > it's fixed frequency (except in idle) irrespective of the frequency
> > the CPU is operating at!
> > it's a "time stamp counter" not a "cpu cycle counter". You can run
> > instructions faster than the tsc increment or slower. Or sometimes
> > at the same rate.
> > Which it is depends on what cpufreq/ondemand are doing and how
> > active IDA is.
> >
>
> I was under the impression that it was the same clock on AMD Opteron
> and newer processors.

for them it's the same: it's independent of the actual cpu clock.
(and constant rate; to be constant rate you pretty much HAVE to be
independent, because you CAN do cpu frequency scaling, even if you as
user decide to not use it)
>
> So if all this is true how and why can the kernel use it but user
> land is wrong for doing so?
>

the kernel doesn't (really) use the frequency (other than reporting it
in /proc). And with tsc it is very very careful.

2008-06-11 20:39:18

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

Mark Hounschell wrote:
> Arjan van de Ven wrote:
>> On Wed, 11 Jun 2008 12:59:24 -0400
>> Mark Hounschell <[email protected]> wrote:
>>
>>> Arjan van de Ven wrote:
>>>> On Wed, 11 Jun 2008 10:14:50 -0400
>>>> "linux-os (Dick Johnson)" <[email protected]> wrote:
>>>>
>>>>> I use CPU frequency to scale the output of rdtsc when timing
>>>>> routines being optimized.
>>>> Ok this is broken in various really bad ways.
>>>>
>>>> 1) CPU frequency is very fluctuating, it changes in principle every
>>>> millisecond or more
>>> I'm curious. What do you mean by "in principle"? And why will it
>>> change?
>> ondemand governor will change the cpu frequency dynamically all the
>> time.
>> the cpu itself has a dynamic range in which it operates (at least on
>> cpus that support Intel Dynamic Acceleration technology, IDA)
>>
>
> OK. If I am always using AMD Opteron/X64 class machines am I safe here
> when no cpu freq scaling stuff is on and I'm pinned to a particular processor?
>
>>>> 2) the rdtsc "frequency" is conceptually unrelated to cpu
>>>> frequency. In fact, you'll be hard-pressed to buy a system today
>>>> where this relationship works....
>>>>
>>> And what do you mean by "conceptually unrelated to cpu frequency"?
>>> Is it not the clock freq that is driving the cpu and the freq at which
>>> the tsc is incremented?
>> no it is not that... at all.
>> the tsc comes from an entirely different clock, and on anything you can
>> buy today from AMD or Intel (or the last year for that matter), it's
>> fixed frequency (except in idle) irrespective of the frequency the CPU
>> is operating at!
>> it's a "time stamp counter" not a "cpu cycle counter". You can run
>> instructions faster than the tsc increment or slower. Or sometimes at
>> the same rate.
>> Which it is depends on what cpufreq/ondemand are doing and how active
>> IDA is.
>>
>
> I was under the impression that it was the same clock on AMD Opteron and
> newer processors.
>
> So if all this is true how and why can the kernel use it but user land is wrong
> for doing so?
>
> Mark
>

Ok, what would one use besides the tsc to take a fast high-res time stamp in a kernel module that
could also be used in user land (same time source) to for instance calculate the
the time between an event in the kernel to an event in user land?

do_gettimeofday / gettimeofday are not really acceptable.

Mark

2008-06-11 20:56:42

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 16:37:38 -0400
> > Mark
> >
>
> Ok, what would one use besides the tsc to take a fast high-res time
> stamp in a kernel module that could also be used in user land (same
> time source) to for instance calculate the the time between an event
> in the kernel to an event in user land?
>
> do_gettimeofday / gettimeofday are not really acceptable.

why not? both of these (they're teh same) are the best the kernel can
do. If there was a better way to do it, gettimeofday() would use it.
It's not even a system call if we can get away with it
(although system calls are pretty cheap)

really, think about it. if there was a better way, gettimeofday() would
use it.

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-11 20:58:43

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

On Wed, 11 Jun 2008 16:37:38 -0400
Mark Hounschell <[email protected]> wrote:

> > Mark
> >
>
> Ok, what would one use besides the tsc to take a fast high-res time
> stamp in a kernel module that could also be used in user land (same
> time source) to for instance calculate the the time between an event
> in the kernel to an event in user land?
>
> do_gettimeofday / gettimeofday are not really acceptable.

btw you also changed the problem from "what is the cpu frequency" to
"how usable is rdtsc for time". My entire point was that these weren't
related...

--
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings,
visit http://www.lesswatts.org

2008-06-11 21:00:46

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

Arjan van de Ven wrote:
> On Wed, 11 Jun 2008 16:37:38 -0400
>>> Mark
>>>
>> Ok, what would one use besides the tsc to take a fast high-res time
>> stamp in a kernel module that could also be used in user land (same
>> time source) to for instance calculate the the time between an event
>> in the kernel to an event in user land?
>>
>> do_gettimeofday / gettimeofday are not really acceptable.
>
>
> why not? both of these (they're teh same) are the best the kernel can
> do. If there was a better way to do it, gettimeofday() would use it.
> It's not even a system call if we can get away with it
> (although system calls are pretty cheap)
>
> really, think about it. if there was a better way, gettimeofday() would
> use it.
>
>

The results of gettimeofday are worse that what you describe about the tsc.
NTP etc... and take considerably longer the a tsc read.

Mark

2008-06-11 21:09:40

[permalink] [raw]

Subject: Re: PROBLEM: no cpu MHz in /proc/cpuinfo on 2.6.25.4-rt6

Arjan van de Ven wrote:
> On Wed, 11 Jun 2008 16:37:38 -0400
> Mark Hounschell <[email protected]> wrote:
>
>>> Mark
>>>
>> Ok, what would one use besides the tsc to take a fast high-res time
>> stamp in a kernel module that could also be used in user land (same
>> time source) to for instance calculate the the time between an event
>> in the kernel to an event in user land?
>>
>> do_gettimeofday / gettimeofday are not really acceptable.
>
>
> btw you also changed the problem from "what is the cpu frequency" to
> "how usable is rdtsc for time". My entire point was that these weren't
> related...
>
>

If they are not the same how do you know at what rate the tsc is counting.

But back to "how usable is rdtsc for time", is there a

clock_gettime(CLOCK_MONOTONIC, &time) equivalent that can be used in the kernel?

Mark

2008-06-11 21:28:57