hardirq.h removal broke sparc64 build...

hardirq.h removal broke sparc64 build...

[David Miller]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: Text/Plain; charset=utf-8, Size: 2606 bytes --]


[ Andrew, read carefully, I think some part of a patch you
  merged might have been lost while reshuffling or whatever. ]

This change:

commit ba84be2338d3a2b6020d39279335bb06fcd332e1
Author: Russell King <rmk+kernel@arm.linux.org.uk>
Date:   Tue Jan 6 14:41:07 2009 -0800

    remove linux/hardirq.h from asm-generic/local.h

broke the build on sparc64:

drivers/base/topology.c: In function ‘show_physical_package_id’:
drivers/base/topology.c:103: error: implicit declaration of function ‘cpu_data’
drivers/base/topology.c:103: error: request for member ‘proc_id’ in something not a structure or union
drivers/base/topology.c: In function ‘show_core_id’:
drivers/base/topology.c:106: error: request for member ‘core_id’ in something not a structure or union

Even though there is a footnote there from Alexey saying that
sparc64 was fixed up :-)

    [adobriyan@gmail.com: fix sparc64]

I've spent most of last night and this morning trying to figure out
a way to fix this, it is very non-trivial.

Adding a simple asm/cpudata.h include into asm/topology_64.h for the
SMP case does not work.

asm/cpudata.h brings in percpu.h which wants SLAB which wants to use
GFP_* values and thus includes gfp.h

And thus because of the include loop:

	linux/gfp.h --> linux/mmzone.h --> linux/topology.h -->
	asm/topology.h --> asm/cpudata.h --> linux/percpu.h -->
	linux/slab.h

we can't include asm/cpudata.h into asm/topology.h otherwise
we get:

include/linux/slub_def.h: In function ‘kmalloc_large’:
include/linux/slub_def.h:209: error: implicit declaration of function ‘__get_free_pages’
include/linux/slub_def.h:209: error: ‘__GFP_COMP’ undeclared (first use in this function)
include/linux/slub_def.h:209: error: (Each undeclared identifier is reported only once
include/linux/slub_def.h:209: error: for each function it appears in.)
include/linux/slub_def.h:209: warning: cast to pointer from integer of different size

for the drivers/base/topology.o build failure case above.

Actually, looking at the patch, where is the part from Alexey
that does the "fix sparc64"?  I see nothing in the patch that
would even appear to possibly be for that.  The diffstat is:

 arch/avr32/kernel/traps.c     |    1 +
 arch/sh/kernel/traps_32.c     |    1 +
 drivers/acpi/processor_idle.c |    1 +
 include/asm-generic/local.h   |    1 -
 4 files changed, 3 insertions(+), 1 deletion(-)

Was Alexey's fix removed or lost? :-/
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: hardirq.h removal broke sparc64 build...

[Andrew Morton]

On Wed, 07 Jan 2009 14:43:10 -0800 (PST)
David Miller <davem@davemloft.net> wrote:

> 
> [ Andrew, read carefully, I think some part of a patch you
>   merged might have been lost while reshuffling or whatever. ]
> 
> This change:
> 
> commit ba84be2338d3a2b6020d39279335bb06fcd332e1
> Author: Russell King <rmk+kernel@arm.linux.org.uk>
> Date:   Tue Jan 6 14:41:07 2009 -0800
> 
>     remove linux/hardirq.h from asm-generic/local.h
> 
> broke the build on sparc64:

urgh, sorry.

> drivers/base/topology.c: In function ___show_physical_package_id___:
> drivers/base/topology.c:103: error: implicit declaration of function ___cpu_data___
> drivers/base/topology.c:103: error: request for member ___proc_id___ in something not a structure or union
> drivers/base/topology.c: In function ___show_core_id___:
> drivers/base/topology.c:106: error: request for member ___core_id___ in something not a structure or union
> 
> Even though there is a footnote there from Alexey saying that
> sparc64 was fixed up :-)
> 
>     [adobriyan@gmail.com: fix sparc64]

OK, that patch was misdescribed.  It in fact added the <asm/hardirq.h>
includes to arch/avr32/kernel/traps.c and arch/sh/kernel/traps_32.c.

The "sparc fix" was Sam's patch:

--- a/arch/sparc/kernel/mdesc.c~remove-linux-hardirqh-from-asm-generic-localh-sparc-fix
+++ a/arch/sparc/kernel/mdesc.c
@@ -11,6 +11,7 @@
 #include <linux/mm.h>
 #include <linux/miscdevice.h>
 
+#include <asm/cpudata.h>
 #include <asm/hypervisor.h>
 #include <asm/mdesc.h>
 #include <asm/prom.h>
--- a/arch/sparc/kernel/sysfs.c~remove-linux-hardirqh-from-asm-generic-localh-sparc-fix
+++ a/arch/sparc/kernel/sysfs.c
@@ -8,6 +8,7 @@
 #include <linux/percpu.h>
 #include <linux/init.h>
 
+#include <asm/cpudata.h>
 #include <asm/hypervisor.h>
 #include <asm/spitfire.h>


which did get included. 


> I've spent most of last night and this morning trying to figure out
> a way to fix this, it is very non-trivial.
> 
> Adding a simple asm/cpudata.h include into asm/topology_64.h for the
> SMP case does not work.
> 
> asm/cpudata.h brings in percpu.h which wants SLAB which wants to use
> GFP_* values and thus includes gfp.h
> 
> And thus because of the include loop:
> 
> 	linux/gfp.h --> linux/mmzone.h --> linux/topology.h -->
> 	asm/topology.h --> asm/cpudata.h --> linux/percpu.h -->
> 	linux/slab.h
> 
> we can't include asm/cpudata.h into asm/topology.h otherwise
> we get:
> 
> include/linux/slub_def.h: In function ___kmalloc_large___:
> include/linux/slub_def.h:209: error: implicit declaration of function _____get_free_pages___
> include/linux/slub_def.h:209: error: _____GFP_COMP___ undeclared (first use in this function)
> include/linux/slub_def.h:209: error: (Each undeclared identifier is reported only once
> include/linux/slub_def.h:209: error: for each function it appears in.)
> include/linux/slub_def.h:209: warning: cast to pointer from integer of different size
> 
> for the drivers/base/topology.o build failure case above.

yes, there were some nasty problems liek that.  In several cases we wimped
out and added the necessary includes to the .c file.  Will that (sad hack)
work here?

Re: hardirq.h removal broke sparc64 build...

[David Miller]

From: Andrew Morton <akpm@linux-foundation.org>
Date: Wed, 7 Jan 2009 14:58:22 -0800

> On Wed, 07 Jan 2009 14:43:10 -0800 (PST)
> David Miller <davem@davemloft.net> wrote:
> 
> > for the drivers/base/topology.o build failure case above.
> 
> yes, there were some nasty problems liek that.  In several cases we wimped
> out and added the necessary includes to the .c file.  Will that (sad hack)
> work here?

I'll try that but if other gremlins pop up I'll simply
add a linux/hardirq.h include to arch/sparc/include/asm/local.h

Re: hardirq.h removal broke sparc64 build...

[David Miller]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: Text/Plain; charset=utf-8, Size: 3103 bytes --]

From: Andrew Morton <akpm@linux-foundation.org>
Date: Wed, 7 Jan 2009 14:58:22 -0800

> On Wed, 07 Jan 2009 14:43:10 -0800 (PST)
> David Miller <davem@davemloft.net> wrote:
> 
> > I've spent most of last night and this morning trying to figure out
> > a way to fix this, it is very non-trivial.
> > 
> > Adding a simple asm/cpudata.h include into asm/topology_64.h for the
> > SMP case does not work.
> > 
> > asm/cpudata.h brings in percpu.h which wants SLAB which wants to use
> > GFP_* values and thus includes gfp.h
> > 
> > And thus because of the include loop:
 ...
> > we can't include asm/cpudata.h into asm/topology.h otherwise
> > we get:
 ...
> > for the drivers/base/topology.o build failure case above.
> 
> yes, there were some nasty problems like that.  In several cases we wimped
> out and added the necessary includes to the .c file.  Will that (sad hack)
> work here?

Ok, this works.

Linus, please apply, thanks!

topology: Fix sparc64 build.

Due to changeset ba84be2338d3a2b6020d39279335bb06fcd332e1
("remove linux/hardirq.h from asm-generic/local.h") the
sparc64 build started failing on drivers/base/topology.c:

drivers/base/topology.c: In function ‘show_physical_package_id’:
drivers/base/topology.c:103: error: implicit declaration of function ‘cpu_data’
drivers/base/topology.c:103: error: request for member ‘proc_id’ in something not a structure or union
drivers/base/topology.c: In function ‘show_core_id’:
drivers/base/topology.c:106: error: request for member ‘core_id’ in something not a structure or union

Adding the obvious fix of including asm/cpudata.h into
asm/topology.h on sparc64 doesn't fix it, in fact it
makes things worse because of the header file dependency
chain:

	linux/gfp.h --> linux/mmzone.h --> linux/topology.h -->
	asm/topology.h --> asm/cpudata.h --> linux/percpu.h -->
	linux/slab.h

which results in:

include/linux/slub_def.h: In function ‘kmalloc_large’:
include/linux/slub_def.h:209: error: implicit declaration of function ‘__get_free_pages’
include/linux/slub_def.h:209: error: ‘__GFP_COMP’ undeclared (first use in this function)
include/linux/slub_def.h:209: error: (Each undeclared identifier is reported only once
include/linux/slub_def.h:209: error: for each function it appears in.)
include/linux/slub_def.h:209: warning: cast to pointer from integer of different size

The simplest thing to do is to add yet another one-off hack like
parts of the guilty changeset did, by putting an explicit
linux/hardirq.h include into drivers/base/topology.c

Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/base/topology.c b/drivers/base/topology.c
index a8bc1cb..a778fb5 100644
--- a/drivers/base/topology.c
+++ b/drivers/base/topology.c
@@ -28,6 +28,7 @@
 #include <linux/mm.h>
 #include <linux/cpu.h>
 #include <linux/module.h>
+#include <linux/hardirq.h>
 #include <linux/topology.h>
 
 #define define_one_ro(_name) 		\
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: hardirq.h removal broke sparc64 build...

[Sam Ravnborg]

> 
> Actually, looking at the patch, where is the part from Alexey
> that does the "fix sparc64"?  I see nothing in the patch that
> would even appear to possibly be for that.  The diffstat is:
> 
>  arch/avr32/kernel/traps.c     |    1 +
>  arch/sh/kernel/traps_32.c     |    1 +
>  drivers/acpi/processor_idle.c |    1 +
>  include/asm-generic/local.h   |    1 -
>  4 files changed, 3 insertions(+), 1 deletion(-)
> 
> Was Alexey's fix removed or lost? :-/

When I made my patch I saw some issues in non-sparc files
but I did not address these as I thought they were generic
bugs addressed by someone else.
I should have checked if the same build error was
also present on x86/arm then :-(

I'm glad to see it is sorted out properly now.

	Sam