Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[David Miller]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: Text/Plain; charset=utf-8, Size: 1586 bytes --]

From: Robert Reif <reif@earthlink.net>
Date: Tue, 13 Jan 2009 20:40:04 -0500

> Todays git produces this compile error on sparc32:
> 
> kernel/up.c:10: error: redefinition of ‘smp_call_function_single’
> /usr/src/sparc32/linux-2.6/arch/sparc/include/asm/smp_32.h:78: error: previous definition of ‘smp_call_function_single’ was here

Andrew, are you sure that Makefile condition for compiling
up.o is correct?

ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
obj-y += smp.o
else
obj-y += up.o
endif

If an arch doesn't use generic SMP helpers, there is no way
we should compile up.c into the tree when SMP.

This build regression was caused by:

commit 53ce3d9564908794ae7dd32969089b57df5fc098
Author: Andrew Morton <akpm@linux-foundation.org>
Date:   Fri Jan 9 12:27:08 2009 -0800

    smp_call_function_single(): be slightly less stupid
    
    If you do
    
    	smp_call_function_single(expression-with-side-effects, ...)
    
    then expression-with-side-effects never gets evaluated on UP builds.
    
    As always, implementing it in C is the correct thing to do.
    
    While we're there, uninline it for size and possible header dependency
    reasons.
    
    And create a new kernel/up.c, as a place in which to put
    uniprocessor-specific code and storage.  It should mirror kernel/smp.c.
    
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
    Signed-off-by: Ingo Molnar <mingo@elte.hu>
ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[Andrew Morton]

On Tue, 13 Jan 2009 18:10:39 -0800 (PST) David Miller <davem@davemloft.net> wrote:

> From: Robert Reif <reif@earthlink.net>
> Date: Tue, 13 Jan 2009 20:40:04 -0500
> 
> > Todays git produces this compile error on sparc32:
> > 
> > kernel/up.c:10: error: redefinition of ___smp_call_function_single___
> > /usr/src/sparc32/linux-2.6/arch/sparc/include/asm/smp_32.h:78: error: previous definition of ___smp_call_function_single___ was here
> 
> Andrew, are you sure that Makefile condition for compiling
> up.o is correct?

you give me grounds for now being ;)

> ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
> obj-y += smp.o
> else
> obj-y += up.o
> endif
> 
> If an arch doesn't use generic SMP helpers, there is no way
> we should compile up.c into the tree when SMP.
> 

How's about this?

--- a/kernel/Makefile~a
+++ a/kernel/Makefile
@@ -43,7 +43,7 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
 ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
 obj-y += smp.o
 else
-obj-y += up.o
+obj-$(CONFIG_SMP) += up.o
 endif
 obj-$(CONFIG_SMP) += spinlock.o
 obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
_

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[Andrew Morton]

On Tue, 13 Jan 2009 18:42:39 -0800 Andrew Morton <akpm@linux-foundation.org> wrote:

> On Tue, 13 Jan 2009 18:10:39 -0800 (PST) David Miller <davem@davemloft.net> wrote:
> 
> > From: Robert Reif <reif@earthlink.net>
> > Date: Tue, 13 Jan 2009 20:40:04 -0500
> > 
> > > Todays git produces this compile error on sparc32:
> > > 
> > > kernel/up.c:10: error: redefinition of ___smp_call_function_single___
> > > /usr/src/sparc32/linux-2.6/arch/sparc/include/asm/smp_32.h:78: error: previous definition of ___smp_call_function_single___ was here
> > 
> > Andrew, are you sure that Makefile condition for compiling
> > up.o is correct?
> 
> you give me grounds for now being ;)
> 
> > ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
> > obj-y += smp.o
> > else
> > obj-y += up.o
> > endif
> > 
> > If an arch doesn't use generic SMP helpers, there is no way
> > we should compile up.c into the tree when SMP.
> > 
> 
> How's about this?
> 
> --- a/kernel/Makefile~a
> +++ a/kernel/Makefile
> @@ -43,7 +43,7 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
>  ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
>  obj-y += smp.o
>  else
> -obj-y += up.o
> +obj-$(CONFIG_SMP) += up.o
>  endif
>  obj-$(CONFIG_SMP) += spinlock.o
>  obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o

err, the other way of course...

--- a/kernel/Makefile~a
+++ a/kernel/Makefile
@@ -43,6 +43,7 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
 ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
 obj-y += smp.o
 else
+ifneq ($(CONFIG_SMP),y)
 obj-y += up.o
 endif
 obj-$(CONFIG_SMP) += spinlock.o
_

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[Robert Reif]

Andrew Morton wrote:
> On Tue, 13 Jan 2009 18:42:39 -0800 Andrew Morton <akpm@linux-foundation.org> wrote:
>
>   
>> On Tue, 13 Jan 2009 18:10:39 -0800 (PST) David Miller <davem@davemloft.net> wrote:
>>
>>     
>>> From: Robert Reif <reif@earthlink.net>
>>> Date: Tue, 13 Jan 2009 20:40:04 -0500
>>>
>>>       
>>>> Todays git produces this compile error on sparc32:
>>>>
>>>> kernel/up.c:10: error: redefinition of ___smp_call_function_single___
>>>> /usr/src/sparc32/linux-2.6/arch/sparc/include/asm/smp_32.h:78: error: previous definition of ___smp_call_function_single___ was here
>>>>         
>>> Andrew, are you sure that Makefile condition for compiling
>>> up.o is correct?
>>>       
>> you give me grounds for now being ;)
>>
>>     
>>> ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
>>> obj-y += smp.o
>>> else
>>> obj-y += up.o
>>> endif
>>>
>>> If an arch doesn't use generic SMP helpers, there is no way
>>> we should compile up.c into the tree when SMP.
>>>
>>>       
>> How's about this?
>>
>> --- a/kernel/Makefile~a
>> +++ a/kernel/Makefile
>> @@ -43,7 +43,7 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
>>  ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
>>  obj-y += smp.o
>>  else
>> -obj-y += up.o
>> +obj-$(CONFIG_SMP) += up.o
>>  endif
>>  obj-$(CONFIG_SMP) += spinlock.o
>>  obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
>>     
>
> err, the other way of course...
>
> --- a/kernel/Makefile~a
> +++ a/kernel/Makefile
> @@ -43,6 +43,7 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
>  ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
>  obj-y += smp.o
>  else
> +ifneq ($(CONFIG_SMP),y)
>  obj-y += up.o
>  endif
>  obj-$(CONFIG_SMP) += spinlock.o
> _
>
>
>
>   
This worked:

diff --git a/kernel/Makefile b/kernel/Makefile
index 2aebc4c..368227d 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -43,8 +43,10 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
 ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
 obj-y += smp.o
 else
+ifneq ($(CONFIG_SMP),y)
 obj-y += up.o
 endif
+endif
 obj-$(CONFIG_SMP) += spinlock.o
 obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
 obj-$(CONFIG_PROVE_LOCKING) += spinlock.o

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[Andrew Morton]

On Tue, 13 Jan 2009 22:38:01 -0500 Robert Reif <reif@earthlink.net> wrote:

> This worked:
> 
> diff --git a/kernel/Makefile b/kernel/Makefile
> index 2aebc4c..368227d 100644
> --- a/kernel/Makefile
> +++ b/kernel/Makefile
> @@ -43,8 +43,10 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
>  ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
>  obj-y += smp.o
>  else
> +ifneq ($(CONFIG_SMP),y)
>  obj-y += up.o
>  endif
> +endif
>  obj-$(CONFIG_SMP) += spinlock.o
>  obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
>  obj-$(CONFIG_PROVE_LOCKING) += spinlock.o

This all can be simplified, can't it?

obj-$(CONFIG_USE_GENERIC_SMP_HELPERS) += smp.o
ifneq ($(CONFIG_SMP),y)
obj-y += up.o
endif

(someone please check my homework - I don't have a good track
record here ;))

--- a/kernel/Makefile~kernel-upc-omit-it-if-smp=y-use_generic_smp_helpers=n
+++ a/kernel/Makefile
@@ -40,9 +40,8 @@ obj-$(CONFIG_RT_MUTEXES) += rtmutex.o
 obj-$(CONFIG_DEBUG_RT_MUTEXES) += rtmutex-debug.o
 obj-$(CONFIG_RT_MUTEX_TESTER) += rtmutex-tester.o
 obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
-ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
-obj-y += smp.o
-else
+obj-$(CONFIG_USE_GENERIC_SMP_HELPERS) += smp.o
+ifneq ($(CONFIG_SMP),y)
 obj-y += up.o
 endif
 obj-$(CONFIG_SMP) += spinlock.o
_

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[Sam Ravnborg]

On Tue, Jan 13, 2009 at 11:59:40PM -0800, Andrew Morton wrote:
> On Tue, 13 Jan 2009 22:38:01 -0500 Robert Reif <reif@earthlink.net> wrote:
> 
> > This worked:
> > 
> > diff --git a/kernel/Makefile b/kernel/Makefile
> > index 2aebc4c..368227d 100644
> > --- a/kernel/Makefile
> > +++ b/kernel/Makefile
> > @@ -43,8 +43,10 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
> >  ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
> >  obj-y += smp.o
> >  else
> > +ifneq ($(CONFIG_SMP),y)
> >  obj-y += up.o
> >  endif
> > +endif
> >  obj-$(CONFIG_SMP) += spinlock.o
> >  obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
> >  obj-$(CONFIG_PROVE_LOCKING) += spinlock.o
> 
> This all can be simplified, can't it?
> 
> obj-$(CONFIG_USE_GENERIC_SMP_HELPERS) += smp.o
> ifneq ($(CONFIG_SMP),y)
> obj-y += up.o
> endif
> 
> (someone please check my homework - I don't have a good track
> record here ;))

Looks correct. We pull in smp.o only for SPARC64 AND SMP

But I find the next lines distastefull in a Makefile:
> ifneq ($(CONFIG_SMP),y)
> obj-y += up.o
> endif

I would prefer a small Kconfig helper symbol:

config SPARC_UP
	def_bool y
	depends on !SMP

And then we would do the Makefile bits like this:
obj-$(CONFIG_SPARC_UP) += up.o

So maybe a bit more involved patch - but more readable IMO.
Robert - can I ask you to fix it up if you and others agree.

Thanks,
	Sam

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[Andrew Morton]

On Wed, 14 Jan 2009 16:32:55 +0100 Sam Ravnborg <sam@ravnborg.org> wrote:

> On Tue, Jan 13, 2009 at 11:59:40PM -0800, Andrew Morton wrote:
> > On Tue, 13 Jan 2009 22:38:01 -0500 Robert Reif <reif@earthlink.net> wrote:
> > 
> > > This worked:
> > > 
> > > diff --git a/kernel/Makefile b/kernel/Makefile
> > > index 2aebc4c..368227d 100644
> > > --- a/kernel/Makefile
> > > +++ b/kernel/Makefile
> > > @@ -43,8 +43,10 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
> > >  ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
> > >  obj-y += smp.o
> > >  else
> > > +ifneq ($(CONFIG_SMP),y)
> > >  obj-y += up.o
> > >  endif
> > > +endif
> > >  obj-$(CONFIG_SMP) += spinlock.o
> > >  obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
> > >  obj-$(CONFIG_PROVE_LOCKING) += spinlock.o
> > 
> > This all can be simplified, can't it?
> > 
> > obj-$(CONFIG_USE_GENERIC_SMP_HELPERS) += smp.o
> > ifneq ($(CONFIG_SMP),y)
> > obj-y += up.o
> > endif
> > 
> > (someone please check my homework - I don't have a good track
> > record here ;))
> 
> Looks correct. We pull in smp.o only for SPARC64 AND SMP

SPARC64=n and SMP, actually (assuming sparc64 is the only
USE_GENERIC_SMP_HELPERS=n arch)

> But I find the next lines distastefull in a Makefile:
> > ifneq ($(CONFIG_SMP),y)
> > obj-y += up.o
> > endif

me too.

> I would prefer a small Kconfig helper symbol:
> 
> config SPARC_UP
> 	def_bool y
> 	depends on !SMP
> 
> And then we would do the Makefile bits like this:
> obj-$(CONFIG_SPARC_UP) += up.o

eek.  Mentioning sparc explicitly in kernel/Makefile is badder.

we could remove zillions of these conditionals if something somewhere
were to generate negated symbols for us.  Say, when kbuild sees
CONFIG_SMP=y, it will generate another symbol: NOT_CONFIG_SMP=y.  So
then we can do

obj-$NOT_CONFIG_SMP += up.o

Or is that too cheesy?

Re: sparc32 compile error: redefinition of ‘smp_call_function_single’

[Sam Ravnborg]

On Wed, Jan 14, 2009 at 10:25:44AM -0800, Andrew Morton wrote:
> On Wed, 14 Jan 2009 16:32:55 +0100 Sam Ravnborg <sam@ravnborg.org> wrote:
> 
> > On Tue, Jan 13, 2009 at 11:59:40PM -0800, Andrew Morton wrote:
> > > On Tue, 13 Jan 2009 22:38:01 -0500 Robert Reif <reif@earthlink.net> wrote:
> > > 
> > > > This worked:
> > > > 
> > > > diff --git a/kernel/Makefile b/kernel/Makefile
> > > > index 2aebc4c..368227d 100644
> > > > --- a/kernel/Makefile
> > > > +++ b/kernel/Makefile
> > > > @@ -43,8 +43,10 @@ obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
> > > >  ifeq ($(CONFIG_USE_GENERIC_SMP_HELPERS),y)
> > > >  obj-y += smp.o
> > > >  else
> > > > +ifneq ($(CONFIG_SMP),y)
> > > >  obj-y += up.o
> > > >  endif
> > > > +endif
> > > >  obj-$(CONFIG_SMP) += spinlock.o
> > > >  obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
> > > >  obj-$(CONFIG_PROVE_LOCKING) += spinlock.o
> > > 
> > > This all can be simplified, can't it?
> > > 
> > > obj-$(CONFIG_USE_GENERIC_SMP_HELPERS) += smp.o
> > > ifneq ($(CONFIG_SMP),y)
> > > obj-y += up.o
> > > endif
> > > 
> > > (someone please check my homework - I don't have a good track
> > > record here ;))
> > 
> > Looks correct. We pull in smp.o only for SPARC64 AND SMP
> 
> SPARC64=n and SMP, actually (assuming sparc64 is the only
> USE_GENERIC_SMP_HELPERS=n arch)
> 
> > But I find the next lines distastefull in a Makefile:
> > > ifneq ($(CONFIG_SMP),y)
> > > obj-y += up.o
> > > endif
> 
> me too.
> 
> > I would prefer a small Kconfig helper symbol:
> > 
> > config SPARC_UP
> > 	def_bool y
> > 	depends on !SMP
> > 
> > And then we would do the Makefile bits like this:
> > obj-$(CONFIG_SPARC_UP) += up.o
> 
> eek.  Mentioning sparc explicitly in kernel/Makefile is badder.

Ups - I my horry I did not notice this was _taht_ kernel/Makefile.

> 
> we could remove zillions of these conditionals if something somewhere
> were to generate negated symbols for us.  Say, when kbuild sees
> CONFIG_SMP=y, it will generate another symbol: NOT_CONFIG_SMP=y.  So
> then we can do
> 
> obj-$NOT_CONFIG_SMP += up.o
> 
> Or is that too cheesy?

If we can remove a zillion lines - then no.
But maybe the actual figure is a bit less :-)

	Sam