Intel x86-64 support patch breaks amd64

Intel x86-64 support patch breaks amd64

[Tony Lindgren]

Hi Andi,

I guess you probably already know about this, but the recent changeset
1.1561.1.1 breaks compiling and booting for amd64.

First, this breaks for the compile:

arch/x86_64/kernel/setup.c: In function detect_ht':
arch/x86_64/kernel/setup.c:599: error: smp_num_siblings' undeclared (first
use in this function)
arch/x86_64/kernel/setup.c:599: error: (Each undeclared identifier is
reported only once
arch/x86_64/kernel/setup.c:599: error: for each function it appears in.)
make[1]: *** [arch/x86_64/kernel/setup.o] Error 1
make: *** [arch/x86_64/kernel] Error 2

After #if 0 out some parts to make it compile, it fails to boot with no
output at all. Sorry, don't have low level debugging or serial console on 
this machine configured, let me know if you need further information.

Undoing this cset makes things work as before.

Regards,

Tony

Re: Intel x86-64 support patch breaks amd64

[Jakub Jelinek]

On Fri, Feb 20, 2004 at 05:13:37PM +0100, Andi Kleen wrote:
> --- linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c-o	2004-02-19 09:01:09.000000000 +0100
> +++ linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c	2004-02-19 09:08:04.000000000 +0100
> @@ -194,7 +194,9 @@
>  
>  EXPORT_SYMBOL(die_chain);
>  
> +#ifdef CONFIG_SMP_

		    ^ Isn't this a typo?

>  EXPORT_SYMBOL(cpu_sibling_map);
> +#endif
>  
>  extern void do_softirq_thunk(void);
>  EXPORT_SYMBOL_NOVERS(do_softirq_thunk);

	Jakub

Re: Intel x86-64 support patch breaks amd64

[Tony Lindgren]

* Andi Kleen <ak@suse.de> [040219 11:24]:
> 
> You need the appended patch to build on Uni Processor again. I already
> submitted it to Linus, but he doesn't seem to have merged it yet
> (or alternatively compile for SMP) 

OK, that compiles, but does not boot. Tt's not the *.S files, not the 
*.c, files, I think it's in the .h files somewhere.

Undoing *.S files did not help. Undoing *.c files did not help. 
Finally undoing the *.h files booted... I'll try to narrow it down more.

Yeah, this is UP machine. Also just remembered I don't have a serial port
for lowe level printk's :)

Tony

Re: Intel x86-64 support patch breaks amd64

[Philippe Elie]

On Fri, 20 Feb 2004 at 17:44 +0000, Andi Kleen wrote:

> > > +#ifdef CONFIG_SMP_
> > 
> > 		    ^ Isn't this a typo?
> 
> Indeed. Thanks for catching it.
> 
> It probably doesn't hurt because I don't know of any module that uses cpu_sibling_map[].
> I think I just copied the export from i386. Maybe it would be best to just remove it completely.

Andrew added it a few hours ago, oprofile use it.

regards,
Phil

Re: Intel x86-64 support patch breaks amd64

[Andi Kleen]

On Thu, 19 Feb 2004 10:34:49 -0800
Tony Lindgren <tony@atomide.com> wrote:

> I guess you probably already know about this, but the recent changeset
> 1.1561.1.1 breaks compiling and booting for amd64.

You need the appended patch to build on Uni Processor again. I already
submitted it to Linus, but he doesn't seem to have merged it yet
(or alternatively compile for SMP) 

> After #if 0 out some parts to make it compile, it fails to boot with no
> output at all. Sorry, don't have low level debugging or serial console on 
> this machine configured, let me know if you need further information.

It works for me with this patch both UP and SMP. Maybe you commented out 
too much? 

-Andi

diff -u linux-2.6.3/arch/x86_64/kernel/setup.c-o linux-2.6.3/arch/x86_64/kernel/setup.c
--- linux-2.6.3/arch/x86_64/kernel/setup.c-o	2004-02-19 09:01:09.000000000 +0100
+++ linux-2.6.3/arch/x86_64/kernel/setup.c	2004-02-19 09:09:27.000000000 +0100
@@ -588,6 +588,7 @@
 
 static void __init detect_ht(void)
 {
+#ifdef CONFIG_SMP
 	extern	int phys_proc_id[NR_CPUS];
 	
 	u32 	eax, ebx, ecx, edx;
@@ -631,6 +632,7 @@
 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
 		       phys_proc_id[cpu]);
 	}
+#endif
 }
 	
 #define LVL_1_INST	1
diff -u linux-2.6.3/arch/x86_64/kernel/Makefile-o linux-2.6.3/arch/x86_64/kernel/Makefile
--- linux-2.6.3/arch/x86_64/kernel/Makefile-o	2004-02-19 09:01:09.000000000 +0100
+++ linux-2.6.3/arch/x86_64/kernel/Makefile	2004-02-19 09:15:41.000000000 +0100
@@ -33,4 +33,4 @@
 cpuid-$(subst m,y,$(CONFIG_X86_CPUID))  += ../../i386/kernel/cpuid.o
 topology-y                     += ../../i386/mach-default/topology.o
 swiotlb-$(CONFIG_SWIOTLB)      += ../../ia64/lib/swiotlb.o
-microcode-$(CONFIG_MICROCODE)  += ../../i386/kernel/microcode.o
+microcode-$(subst m,y,$(CONFIG_X86_CPUID))  += ../../i386/kernel/microcode.o
diff -u linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c-o linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c
--- linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c-o	2004-02-19 09:01:09.000000000 +0100
+++ linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c	2004-02-19 09:08:04.000000000 +0100
@@ -194,7 +194,9 @@
 
 EXPORT_SYMBOL(die_chain);
 
+#ifdef CONFIG_SMP_
 EXPORT_SYMBOL(cpu_sibling_map);
+#endif
 
 extern void do_softirq_thunk(void);
 EXPORT_SYMBOL_NOVERS(do_softirq_thunk);

Re: Intel x86-64 support patch breaks amd64

[Andi Kleen]

On Thu, 19 Feb 2004 14:36:10 -0500
Jakub Jelinek <jakub@redhat.com> wrote:

> On Fri, Feb 20, 2004 at 05:13:37PM +0100, Andi Kleen wrote:
> > --- linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c-o	2004-02-19 09:01:09.000000000 +0100
> > +++ linux-2.6.3/arch/x86_64/kernel/x8664_ksyms.c	2004-02-19 09:08:04.000000000 +0100
> > @@ -194,7 +194,9 @@
> >  
> >  EXPORT_SYMBOL(die_chain);
> >  
> > +#ifdef CONFIG_SMP_
> 
> 		    ^ Isn't this a typo?

Indeed. Thanks for catching it.

It probably doesn't hurt because I don't know of any module that uses cpu_sibling_map[].
I think I just copied the export from i386. Maybe it would be best to just remove it completely.

-Andi

Re: Intel x86-64 support patch breaks amd64

[Andi Kleen]

On Thu, 19 Feb 2004 12:39:20 -0800
Tony Lindgren <tony@atomide.com> wrote:

> * Andi Kleen <ak@suse.de> [040219 11:24]:
> > 
> > You need the appended patch to build on Uni Processor again. I already
> > submitted it to Linus, but he doesn't seem to have merged it yet
> > (or alternatively compile for SMP) 
> 
> OK, that compiles, but does not boot. Tt's not the *.S files, not the 
> *.c, files, I think it's in the .h files somewhere.

Not sure what you mean with that.

Can you please fetch a completely fresh tree, apply the patch and try again?

Most likely you have a broken build of some sort.

-Andi

[PATCH] Fix typo in x86-64 fix

[Andi Kleen]

On Thu, 19 Feb 2004 21:20:02 +0000
Philippe Elie <phil.el@wanadoo.fr> wrote:

> On Fri, 20 Feb 2004 at 17:44 +0000, Andi Kleen wrote:
> 
> > > > +#ifdef CONFIG_SMP_
> > > 
> > > 		    ^ Isn't this a typo?
> > 
> > Indeed. Thanks for catching it.
> > 
> > It probably doesn't hurt because I don't know of any module that uses cpu_sibling_map[].
> > I think I just copied the export from i386. Maybe it would be best to just remove it completely.
> 
> Andrew added it a few hours ago, oprofile use it.

Ok, Andrew, can you please apply this incremental patch to fix the typo too ? 

diff -u linux-2.6.3-amd64/arch/x86_64/kernel/x8664_ksyms.c-o linux-2.6.3-amd64/arch/x86_64/kernel/x8664_ksyms.c
--- linux-2.6.3-amd64/arch/x86_64/kernel/x8664_ksyms.c-o	2004-02-20 15:53:53.000000000 +0100
+++ linux-2.6.3-amd64/arch/x86_64/kernel/x8664_ksyms.c	2004-02-20 20:25:03.000000000 +0100
@@ -194,7 +194,7 @@
 
 EXPORT_SYMBOL(die_chain);
 
-#ifdef CONFIG_SMP_
+#ifdef CONFIG_SMP
 EXPORT_SYMBOL(cpu_sibling_map);
 #endif
 

-Andi