linux-2.6.0-test6 error during linking

linux-2.6.0-test6 error during linking

[Paolo Dovera]

Hello,
   I successfully compiled all linux-2.6.0-testx before test6 on RH7.2 
Alpha APX platform

Applied patch-2.6.0-test6 I receive the following error:

[root@alpha linux-2.6.0-test6]# make
make[1]: `arch/alpha/kernel/asm-offsets.s' is up to date.
 CHK     include/linux/compile.h
 GEN     .version
 CHK     include/linux/compile.h
 UPD     include/linux/compile.h
 CC      init/version.o
 LD      init/built-in.o
 LD      .tmp_vmlinux1
kernel/built-in.o(.text+0x480): In function `try_to_wake_up':
: undefined reference to `sched_clock'
kernel/built-in.o(.text+0x484): In function `try_to_wake_up':
: undefined reference to `sched_clock'
kernel/built-in.o(.text+0x140c): In function `schedule':
: undefined reference to `sched_clock'
kernel/built-in.o(.text+0x1410): In function `schedule':
: undefined reference to `sched_clock'
kernel/built-in.o(.text+0x5068): In function `copy_process':
: undefined reference to `sched_clock'
kernel/built-in.o(.text+0x506c): more undefined references to 
`sched_clock' follow
make: *** [.tmp_vmlinux1] Error 1


Do you have some suggestions?

Bye

Re: linux-2.6.0-test6 error during linking

[Falk Hueffner]

Paolo Dovera <pdovera@bmind.it> writes:

> Applied patch-2.6.0-test6 I receive the following error:
> 
> kernel/built-in.o(.text+0x5068): In function `copy_process':
> : undefined reference to `sched_clock'
> kernel/built-in.o(.text+0x506c): more undefined references to
> `sched_clock' follow

sched_clock is not implemented yet for Alpha. It is supposed to return
a high resolution timer. As workaround, you could just copy the
implementation from ARM, which simply returns the value based on
jiffies.  Ideally, one would take into account the cycle counter, but
it wraps every few seconds, which has to be taken care of.

-- 
	Falk

Re: linux-2.6.0-test6 error during linking

[Balint Cristian]

On Sunday 28 September 2003 23:51, Falk Hueffner wrote:
> Paolo Dovera <pdovera@bmind.it> writes:
> > Applied patch-2.6.0-test6 I receive the following error:
> >
> > kernel/built-in.o(.text+0x5068): In function `copy_process':
> > : undefined reference to `sched_clock'
> >
> > kernel/built-in.o(.text+0x506c): more undefined references to
> > `sched_clock' follow
>
> sched_clock is not implemented yet for Alpha. It is supposed to return
> a high resolution timer. As workaround, you could just copy the
> implementation from ARM, which simply returns the value based on
> jiffies.  Ideally, one would take into account the cycle counter, but
> it wraps every few seconds, which has to be taken care of.

Can post here a patch based on arm ?

I am courious to see it !

Eventualy post it on linux kernel list or to Richard Henderson ?

Thx a lot !

Re: linux-2.6.0-test6 error during linking

[Falk Hueffner]

Balint Cristian <rezso@rdsor.ro> writes:

> On Sunday 28 September 2003 23:51, Falk Hueffner wrote:
> > sched_clock is not implemented yet for Alpha. It is supposed to
> > return a high resolution timer. As workaround, you could just copy
> > the implementation from ARM, which simply returns the value based
> > on jiffies.  Ideally, one would take into account the cycle
> > counter, but it wraps every few seconds, which has to be taken
> > care of.
> 
> Can post here a patch based on arm ?

Well, just copy the function:

--- arch/alpha/kernel/time.c~   2003-09-28 02:50:20.000000000 +0200
+++ arch/alpha/kernel/time.c    2003-09-28 13:39:58.000000000 +0200
@@ -570,3 +570,12 @@
 
        return retval;
 }
+
+/*
+ * Scheduler clock - returns current time in nanosec units.
+ * Should probably take rpcc into account for improved resolution
+ */
+unsigned long long sched_clock(void)
+{
+       return (unsigned long long) jiffies * (1000000000 / HZ);
+}

I've not tested this, it should work, but probably scheduling
behaviour will not be as good as it could be, although HZ being 1000
(or larger) on Alpha should help.

> Eventualy post it on linux kernel list or to Richard Henderson ?

Not without testing it, and I have no time for that currently.

-- 
	Falk

Re: linux-2.6.0-test6 error during linking

[Balint Cristian]

On Monday 29 September 2003 12:02, Falk Hueffner wrote:
> Balint Cristian <rezso@rdsor.ro> writes:
> > On Sunday 28 September 2003 23:51, Falk Hueffner wrote:
> > > sched_clock is not implemented yet for Alpha. It is supposed to
> > > return a high resolution timer. As workaround, you could just copy
> > > the implementation from ARM, which simply returns the value based
> > > on jiffies.  Ideally, one would take into account the cycle
> > > counter, but it wraps every few seconds, which has to be taken
> > > care of.
> >
> > Can post here a patch based on arm ?

Thx a lot !

I test it today on an miata machine, i tell the result on that list.


>
> Well, just copy the function:
>
> --- arch/alpha/kernel/time.c~   2003-09-28 02:50:20.000000000 +0200
> +++ arch/alpha/kernel/time.c    2003-09-28 13:39:58.000000000 +0200
> @@ -570,3 +570,12 @@
>
>         return retval;
>  }
> +
> +/*
> + * Scheduler clock - returns current time in nanosec units.
> + * Should probably take rpcc into account for improved resolution
> + */
> +unsigned long long sched_clock(void)
> +{
> +       return (unsigned long long) jiffies * (1000000000 / HZ);
> +}
>
> I've not tested this, it should work, but probably scheduling
> behaviour will not be as good as it could be, although HZ being 1000
> (or larger) on Alpha should help.
>
> > Eventualy post it on linux kernel list or to Richard Henderson ?
>
> Not without testing it, and I have no time for that currently.

Re: linux-2.6.0-test6 error during linking

[Paolo Dovera]

Thanks^2   to you!
It works fine
I'm trying 2.6.0-test6 on Tsunami Alpha platform

Ciao,
Paolo

Falk Hueffner wrote:

>Balint Cristian <rezso@rdsor.ro> writes:
>  
>
>>On Sunday 28 September 2003 23:51, Falk Hueffner wrote:
>>    
>>
>>>sched_clock is not implemented yet for Alpha. It is supposed to
>>>return a high resolution timer. As workaround, you could just copy
>>>the implementation from ARM, which simply returns the value based
>>>on jiffies.  Ideally, one would take into account the cycle
>>>counter, but it wraps every few seconds, which has to be taken
>>>care of.
>>>      
>>>
>>Can post here a patch based on arm ?
>>    
>>
>
>Well, just copy the function:
>
>--- arch/alpha/kernel/time.c~   2003-09-28 02:50:20.000000000 +0200
>+++ arch/alpha/kernel/time.c    2003-09-28 13:39:58.000000000 +0200
>@@ -570,3 +570,12 @@
> 
>        return retval;
> }
>+
>+/*
>+ * Scheduler clock - returns current time in nanosec units.
>+ * Should probably take rpcc into account for improved resolution
>+ */
>+unsigned long long sched_clock(void)
>+{
>+       return (unsigned long long) jiffies * (1000000000 / HZ);
>+}
>
>I've not tested this, it should work, but probably scheduling
>behaviour will not be as good as it could be, although HZ being 1000
>(or larger) on Alpha should help.
>
>  
>
>>Eventualy post it on linux kernel list or to Richard Henderson ?
>>    
>>
>
>Not without testing it, and I have no time for that currently.
>
>  
>

RE: linux-2.6.0-test6 error during linking

[Oliver Pitzeier]

> On Sunday 28 September 2003 23:51, Falk Hueffner wrote:
> > Paolo Dovera <pdovera@bmind.it> writes:
> > > Applied patch-2.6.0-test6 I receive the following error:
> > >
> > > kernel/built-in.o(.text+0x5068): In function `copy_process':
> > > : undefined reference to `sched_clock'
> > >
> > > kernel/built-in.o(.text+0x506c): more undefined references to 
> > > `sched_clock' follow
> >
> > sched_clock is not implemented yet for Alpha. It is 
> supposed to return 
> > a high resolution timer. As workaround, you could just copy the 
> > implementation from ARM, which simply returns the value based on 
> > jiffies.  Ideally, one would take into account the cycle 
> counter, but 
> > it wraps every few seconds, which has to be taken care of.
> 
> Can post here a patch based on arm ?
> 
> I am courious to see it !
> 
> Eventualy post it on linux kernel list or to Richard Henderson ?
> 
> Thx a lot !

Please take a look at the LKML. I had the same problem with -test6 and viro
already posted a workaround!

Best,
 Oliver