2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Gregoire Favre]

Hello,

can't compil it :

  CHK     include/linux/compile.h
  UPD     include/linux/compile.h
  CC      init/version.o
  LD      init/built-in.o
  LD      .tmp_vmlinux1
arch/x86_64/kernel/built-in.o: In function `vsyscall_set_cpu':
(.init.text+0x1e87): undefined reference to `smp_call_function_single'
make: *** [.tmp_vmlinux1] Error 1

Please CC to me if some more infos are needed as I am not on the list.
-- 
Grégoire FAVRE  http://gregoire.favre.googlepages.com  http://www.gnupg.org

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Adrian Bunk]

On Sun, Jul 09, 2006 at 01:49:25PM +0200, Gregoire Favre wrote:
> Hello,
> 
> can't compil it :
> 
>   CHK     include/linux/compile.h
>   UPD     include/linux/compile.h
>   CC      init/version.o
>   LD      init/built-in.o
>   LD      .tmp_vmlinux1
> arch/x86_64/kernel/built-in.o: In function `vsyscall_set_cpu':
> (.init.text+0x1e87): undefined reference to `smp_call_function_single'
> make: *** [.tmp_vmlinux1] Error 1

Thanks for your report, it seems x86_64-mm-getcpu-vsyscall.patch broke 
COMFIG_SMP=n x86_64 compiles.

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Gregoire Favre]

On Sun, Jul 09, 2006 at 03:00:27PM +0200, Adrian Bunk wrote:

> Thanks for your report, it seems x86_64-mm-getcpu-vsyscall.patch broke 
> COMFIG_SMP=n x86_64 compiles.

Thanks you very much : reverting x86_64-mm-getcpu-vsyscall.patch I can
compil it again :-)
-- 
Grégoire FAVRE  http://gregoire.favre.googlepages.com  http://www.gnupg.org

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Rafael J. Wysocki]

On Sunday 09 July 2006 13:49, Gregoire Favre wrote:
> Hello,
> 
> can't compil it :
> 
>   CHK     include/linux/compile.h
>   UPD     include/linux/compile.h
>   CC      init/version.o
>   LD      init/built-in.o
>   LD      .tmp_vmlinux1
> arch/x86_64/kernel/built-in.o: In function `vsyscall_set_cpu':
> (.init.text+0x1e87): undefined reference to `smp_call_function_single'
> make: *** [.tmp_vmlinux1] Error 1

This is because of x86_64-mm-getcpu-vsyscall.patch which breaks
compilation without SMP and is not obviously fixable.

I think you can safely revert it.

Greetings,
Rafael

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Andrew Morton]

On Sun, 9 Jul 2006 22:39:48 +0200
"Rafael J. Wysocki" <rjw@sisk.pl> wrote:

> On Sunday 09 July 2006 13:49, Gregoire Favre wrote:
> > Hello,
> > 
> > can't compil it :
> > 
> >   CHK     include/linux/compile.h
> >   UPD     include/linux/compile.h
> >   CC      init/version.o
> >   LD      init/built-in.o
> >   LD      .tmp_vmlinux1
> > arch/x86_64/kernel/built-in.o: In function `vsyscall_set_cpu':
> > (.init.text+0x1e87): undefined reference to `smp_call_function_single'
> > make: *** [.tmp_vmlinux1] Error 1
> 
> This is because of x86_64-mm-getcpu-vsyscall.patch which breaks
> compilation without SMP and is not obviously fixable.

Is it not as simple as adding a !SMP implementation of
smp_call_function_single(), which just calls the thing?

> I think you can safely revert it.

That works, too.

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Rafael J. Wysocki]

On Sunday 09 July 2006 23:03, Andrew Morton wrote:
> On Sun, 9 Jul 2006 22:39:48 +0200
> "Rafael J. Wysocki" <rjw@sisk.pl> wrote:
> 
> > On Sunday 09 July 2006 13:49, Gregoire Favre wrote:
> > > Hello,
> > > 
> > > can't compil it :
> > > 
> > >   CHK     include/linux/compile.h
> > >   UPD     include/linux/compile.h
> > >   CC      init/version.o
> > >   LD      init/built-in.o
> > >   LD      .tmp_vmlinux1
> > > arch/x86_64/kernel/built-in.o: In function `vsyscall_set_cpu':
> > > (.init.text+0x1e87): undefined reference to `smp_call_function_single'
> > > make: *** [.tmp_vmlinux1] Error 1
> > 
> > This is because of x86_64-mm-getcpu-vsyscall.patch which breaks
> > compilation without SMP and is not obviously fixable.
> 
> Is it not as simple as adding a !SMP implementation of
> smp_call_function_single(), which just calls the thing?

Like this:

 arch/x86_64/kernel/vsyscall.c |    6 ++++++
 1 files changed, 6 insertions(+)

Index: linux-2.6.18-rc1-mm1/arch/x86_64/kernel/vsyscall.c
===================================================================
--- linux-2.6.18-rc1-mm1.orig/arch/x86_64/kernel/vsyscall.c
+++ linux-2.6.18-rc1-mm1/arch/x86_64/kernel/vsyscall.c
@@ -241,10 +241,12 @@ static ctl_table kernel_root_table2[] = 
 
 #endif
 
+#ifdef CONFIG_SMP
 static void __cpuinit write_rdtscp_cb(void *info)
 {
 	write_rdtscp_aux((unsigned long)info);
 }
+#endif
 
 void __cpuinit vsyscall_set_cpu(int cpu)
 {
@@ -254,10 +256,14 @@ void __cpuinit vsyscall_set_cpu(int cpu)
 	node = cpu_to_node[cpu];
 #endif
 	if (cpu_has(&cpu_data[cpu], X86_FEATURE_RDTSCP)) {
+#ifdef CONFIG_SMP
 		/* the notifier is unfortunately not executed on the
 		   target CPU */
 		void *info = (void *)((node << 12) | cpu);
 		smp_call_function_single(cpu, write_rdtscp_cb, info, 0, 1);
+#else
+		write_rdtscp_aux(0);
+#endif
 	}
 
 	/* Store cpu number in limit so that it can be loaded quickly

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Andrew Morton]

On Mon, 10 Jul 2006 00:37:14 +0200
"Rafael J. Wysocki" <rjw@sisk.pl> wrote:

> On Sunday 09 July 2006 23:03, Andrew Morton wrote:
> > On Sun, 9 Jul 2006 22:39:48 +0200
> > "Rafael J. Wysocki" <rjw@sisk.pl> wrote:
> > 
> > > On Sunday 09 July 2006 13:49, Gregoire Favre wrote:
> > > > Hello,
> > > > 
> > > > can't compil it :
> > > > 
> > > >   CHK     include/linux/compile.h
> > > >   UPD     include/linux/compile.h
> > > >   CC      init/version.o
> > > >   LD      init/built-in.o
> > > >   LD      .tmp_vmlinux1
> > > > arch/x86_64/kernel/built-in.o: In function `vsyscall_set_cpu':
> > > > (.init.text+0x1e87): undefined reference to `smp_call_function_single'
> > > > make: *** [.tmp_vmlinux1] Error 1
> > > 
> > > This is because of x86_64-mm-getcpu-vsyscall.patch which breaks
> > > compilation without SMP and is not obviously fixable.
> > 
> > Is it not as simple as adding a !SMP implementation of
> > smp_call_function_single(), which just calls the thing?
> 
> Like this:
> 
>  arch/x86_64/kernel/vsyscall.c |    6 ++++++
>  1 files changed, 6 insertions(+)
> 
> Index: linux-2.6.18-rc1-mm1/arch/x86_64/kernel/vsyscall.c
> ===================================================================
> --- linux-2.6.18-rc1-mm1.orig/arch/x86_64/kernel/vsyscall.c
> +++ linux-2.6.18-rc1-mm1/arch/x86_64/kernel/vsyscall.c
> @@ -241,10 +241,12 @@ static ctl_table kernel_root_table2[] = 
>  
>  #endif
>  
> +#ifdef CONFIG_SMP
>  static void __cpuinit write_rdtscp_cb(void *info)
>  {
>  	write_rdtscp_aux((unsigned long)info);
>  }
> +#endif
>  
>  void __cpuinit vsyscall_set_cpu(int cpu)
>  {
> @@ -254,10 +256,14 @@ void __cpuinit vsyscall_set_cpu(int cpu)
>  	node = cpu_to_node[cpu];
>  #endif
>  	if (cpu_has(&cpu_data[cpu], X86_FEATURE_RDTSCP)) {
> +#ifdef CONFIG_SMP
>  		/* the notifier is unfortunately not executed on the
>  		   target CPU */
>  		void *info = (void *)((node << 12) | cpu);
>  		smp_call_function_single(cpu, write_rdtscp_cb, info, 0, 1);
> +#else
> +		write_rdtscp_aux(0);
> +#endif
>  	}
>  
>  	/* Store cpu number in limit so that it can be loaded quickly

I meant, in smp.h:

#else	/* CONFIG_SMP */
#define smp_call_function_single(cpu, fn, arg, x, y) fn(arg)
#endif	/* CONFIG_SMP */

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Andi Kleen]

> > Like this:
> > 
> >  arch/x86_64/kernel/vsyscall.c |    6 ++++++
> >  1 files changed, 6 insertions(+)

Sorry will do a proper patch.

-Andi

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Chuck Ebbert]

In-Reply-To: <20060709154445.60d6619c.akpm@osdl.org>

On Sun, 9 Jul 2006 15:44:45 -0700, Andrew Morton wrote:

> I meant, in smp.h:
> 
> #else /* CONFIG_SMP */
> #define smp_call_function_single(cpu, fn, arg, x, y) fn(arg)
> #endif        /* CONFIG_SMP */

But smp_call_function_single() generates an error if you try to call
it on your own CPU, so that doesn't make sense.

I fixed it like this, because that register defaults to zero
anyway and doesn't need initialization on CPU 0.

What I can't figure out is how this ever gets called on CPU 0
during init, whether it's SMP or not.

Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com>

--- 2.6.18-rc1-mm1-64.orig/arch/x86_64/kernel/vsyscall.c
+++ 2.6.18-rc1-mm1-64/arch/x86_64/kernel/vsyscall.c
@@ -253,13 +253,14 @@ void __cpuinit vsyscall_set_cpu(int cpu)
 #ifdef CONFIG_NUMA
 	node = cpu_to_node[cpu];
 #endif
+#ifdef CONFIG_SMP
 	if (cpu_has(&cpu_data[cpu], X86_FEATURE_RDTSCP)) {
 		/* the notifier is unfortunately not executed on the
 		   target CPU */
 		void *info = (void *)((node << 12) | cpu);
 		smp_call_function_single(cpu, write_rdtscp_cb, info, 0, 1);
 	}
-
+#endif
 	/* Store cpu number in limit so that it can be loaded quickly
 	   in user space in vgetcpu.
 	   12 bits for the CPU and 8 bits for the node. */
-- 
Chuck
 "You can't read a newspaper if you can't read."  --George W. Bush

Re: 2.6.18-rc1-mm1 fails on amd64 (smp_call_function_single)

[Andi Kleen]

On Monday 10 July 2006 03:49, Chuck Ebbert wrote:
> In-Reply-To: <20060709154445.60d6619c.akpm@osdl.org>
> 
> On Sun, 9 Jul 2006 15:44:45 -0700, Andrew Morton wrote:
> 
> > I meant, in smp.h:
> > 
> > #else /* CONFIG_SMP */
> > #define smp_call_function_single(cpu, fn, arg, x, y) fn(arg)
> > #endif        /* CONFIG_SMP */
> 
> But smp_call_function_single() generates an error if you try to call
> it on your own CPU, so that doesn't make sense.

I have a full patch to be mirrored out soon.

Your patch is still wrong because now it won't be initialized on the BP

> 
> I fixed it like this, because that register defaults to zero
> anyway and doesn't need initialization on CPU 0.
> 
> What I can't figure out is how this ever gets called on CPU 0
> during init, whether it's SMP or not.

The notifier is called from time.c

-Andi