[PATCH] tile: remove a non-existing header

[PATCH] tile: remove a non-existing header

[Mauro Carvalho Chehab]

While testing some API changes at the EDAC core, I've tested the EDAC
drivers compilation on all supported architectures, including Tilera.

I noticed that Tilera upstream code has issues: it doesn't compile with
kernel 3.2:

In file included from include/linux/linkage.h:5,
                 from include/linux/kernel.h:13,
                 from include/linux/sched.h:55,
                 from arch/tile/kernel/asm-offsets.c:19:
/home/mchehab/redhat/trees/work/new_edac/arch/tile/include/asm/linkage.h:18:22: error: feedback.h: No such file or directory
make[1]: *** [arch/tile/kernel/asm-offsets.s] Error 1

Without it, it is not possible to compile anything on this archictecture
using this default configuration:
	./arch/tile/configs/tilegx_defconfig

Cc: stable@kernel.org
Signed-of-by: Mauro Carvalho Chehab <mchehab@redhat.com>

---

Btw, the ./arch/tile/configs/tilepro_defconfig doesn't compile even
with this fix. It fails at arch/tile/include/asm/atomic_64.h because
atomic64_t is never defined, but it is used there and on other places.

Also, while the edac_tile driver seems to be for both Tile GX and Tilepro
sub-archs, it is not included on tilegx_defconfig. It would make sense
to add it there, in order to make easier for people to test/use the
EDAC drivers.

---
 arch/tile/include/asm/linkage.h |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)

diff --git a/arch/tile/include/asm/linkage.h b/arch/tile/include/asm/linkage.h
index e121c39..43b298e 100644
--- a/arch/tile/include/asm/linkage.h
+++ b/arch/tile/include/asm/linkage.h
@@ -15,8 +15,6 @@
 #ifndef _ASM_TILE_LINKAGE_H
 #define _ASM_TILE_LINKAGE_H
 
-#include <feedback.h>
-
 #define __ALIGN .align 8
 
 /*
-- 
1.7.8

Re: [PATCH] tile: remove a non-existing header

[Chris Metcalf]

On 3/12/2012 10:18 AM, Mauro Carvalho Chehab wrote:
> While testing some API changes at the EDAC core, I've tested the EDAC
> drivers compilation on all supported architectures, including Tilera.
>
> I noticed that Tilera upstream code has issues: it doesn't compile with
> kernel 3.2:
>
> In file included from include/linux/linkage.h:5,
>                  from include/linux/kernel.h:13,
>                  from include/linux/sched.h:55,
>                  from arch/tile/kernel/asm-offsets.c:19:
> /home/mchehab/redhat/trees/work/new_edac/arch/tile/include/asm/linkage.h:18:22: error: feedback.h: No such file or directory
> make[1]: *** [arch/tile/kernel/asm-offsets.s] Error 1
>
> Without it, it is not possible to compile anything on this archictecture
> using this default configuration:
> 	./arch/tile/configs/tilegx_defconfig
>
> Cc: stable@kernel.org
> Signed-of-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Thank you, but nack.  The header is provided in the latest gcc tip code. 
Sorry for the API thrash!  It's certainly OK to comment it out for the time
being; it just provides some assembly macros for specifying feedback-guided
optimization.

> Btw, the ./arch/tile/configs/tilepro_defconfig doesn't compile even
> with this fix. It fails at arch/tile/include/asm/atomic_64.h because
> atomic64_t is never defined, but it is used there and on other places.

I'll take a look at this; I've been focusing primarily on the tilegx
architecture recently and something may have bitrotted in the tilepro
support.  Thanks for the pointer.

> Also, while the edac_tile driver seems to be for both Tile GX and Tilepro
> sub-archs, it is not included on tilegx_defconfig. It would make sense
> to add it there, in order to make easier for people to test/use the
> EDAC drivers.

That's probably a good idea; we do enable it by default in our actual
shipping configuration.  I've heard mixed things about the utility of
including "config" files in the kernel, so I haven't been bothering to
update them, but it's probably worth a refresh at this point.

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com

Re: [PATCH] tile: remove a non-existing header

[Mauro Carvalho Chehab]

Em 12-03-2012 11:31, Chris Metcalf escreveu:
> On 3/12/2012 10:18 AM, Mauro Carvalho Chehab wrote:
>> While testing some API changes at the EDAC core, I've tested the EDAC
>> drivers compilation on all supported architectures, including Tilera.
>>
>> I noticed that Tilera upstream code has issues: it doesn't compile with
>> kernel 3.2:
>>
>> In file included from include/linux/linkage.h:5,
>>                  from include/linux/kernel.h:13,
>>                  from include/linux/sched.h:55,
>>                  from arch/tile/kernel/asm-offsets.c:19:
>> /home/mchehab/redhat/trees/work/new_edac/arch/tile/include/asm/linkage.h:18:22: error: feedback.h: No such file or directory
>> make[1]: *** [arch/tile/kernel/asm-offsets.s] Error 1
>>
>> Without it, it is not possible to compile anything on this archictecture
>> using this default configuration:
>> 	./arch/tile/configs/tilegx_defconfig
>>
>> Cc: stable@kernel.org
>> Signed-of-by: Mauro Carvalho Chehab <mchehab@redhat.com>
> 
> Thank you, but nack.  The header is provided in the latest gcc tip code. 

Hmm... the gcc cross-compiler I've downloaded from Tilera site about one month
ago didn't have it (tilegx-unknown-linux-gnu-gcc (GCC) 4.4.6).

> Sorry for the API thrash!  It's certainly OK to comment it out for the time
> being; it just provides some assembly macros for specifying feedback-guided
> optimization.

It would be a good idea then to add some Makefile magic to check it,
maybe creating an empty linkage.h if the header is not found at the gcc
toolchain, in order to support all gcc variants that support tile arch.

>> Btw, the ./arch/tile/configs/tilepro_defconfig doesn't compile even
>> with this fix. It fails at arch/tile/include/asm/atomic_64.h because
>> atomic64_t is never defined, but it is used there and on other places.
> 
> I'll take a look at this; I've been focusing primarily on the tilegx
> architecture recently and something may have bitrotted in the tilepro
> support.  Thanks for the pointer.

Anytime.

>> Also, while the edac_tile driver seems to be for both Tile GX and Tilepro
>> sub-archs, it is not included on tilegx_defconfig. It would make sense
>> to add it there, in order to make easier for people to test/use the
>> EDAC drivers.
> 
> That's probably a good idea; we do enable it by default in our actual
> shipping configuration.  I've heard mixed things about the utility of
> including "config" files in the kernel, so I haven't been bothering to
> update them, but it's probably worth a refresh at this point.

Yeah, there are lots of discussion about keeping/removing the config 
files in kernel ;)

I wouldn't be using any of those, if a "make allyesconfig" would 
enable all EDAC drivers for the non-Intel archs, but it didn't work 
like that.

As it didn't enabled, the config files there helped me a lot to test 
compilation for the tile_edac driver, and for some other ppc EDAC 
drivers for embedded platforms. The dependency chain to enable 
those drivers is not trivial, so, trying to enable them by hand 
is a time-consuming task.

Regards,
Mauro

Re: [PATCH] tile: remove a non-existing header

[Chris Metcalf]

On 3/12/2012 11:00 AM, Mauro Carvalho Chehab wrote:
> Em 12-03-2012 11:31, Chris Metcalf escreveu:
>> On 3/12/2012 10:18 AM, Mauro Carvalho Chehab wrote:
>>> While testing some API changes at the EDAC core, I've tested the EDAC
>>> drivers compilation on all supported architectures, including Tilera.
>>>
>>> I noticed that Tilera upstream code has issues: it doesn't compile with
>>> kernel 3.2:
>>>
>>> [...]
>> Thank you, but nack.  The header is provided in the latest gcc tip code. 
> Hmm... the gcc cross-compiler I've downloaded from Tilera site about one month
> ago didn't have it (tilegx-unknown-linux-gnu-gcc (GCC) 4.4.6).

Yes, I need to update it.  We've been focusing on the community return and
haven't updated our own web site recently :-)

>> Sorry for the API thrash!  It's certainly OK to comment it out for the time
>> being; it just provides some assembly macros for specifying feedback-guided
>> optimization.
> It would be a good idea then to add some Makefile magic to check it,
> maybe creating an empty linkage.h if the header is not found at the gcc
> toolchain, in order to support all gcc variants that support tile arch.

True, although, since there hasn't been an official release of gcc that
supports tile yet (the first will be 4.7), I'm less worried about
compatibility issues.  The gcc we ship to our customers has supported it
since day one, and 4.7 and following will support it, so it's just the
Tilera web site version, and I think it's OK for me to just update that one.

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com

Re: [PATCH] tile: remove a non-existing header

[Chris Metcalf]

(Cc's trimmed a bit.)

On 3/12/2012 11:00 AM, Mauro Carvalho Chehab wrote:
> Em 12-03-2012 11:31, Chris Metcalf escreveu:
>> On 3/12/2012 10:18 AM, Mauro Carvalho Chehab wrote:
>>> While testing some API changes at the EDAC core, I've tested the EDAC
>>> drivers compilation on all supported architectures, including Tilera.
>>>
>>> I noticed that Tilera upstream code has issues: it doesn't compile with
>>> kernel 3.2:
>>>
>>> In file included from include/linux/linkage.h:5,
>>>                  from include/linux/kernel.h:13,
>>>                  from include/linux/sched.h:55,
>>>                  from arch/tile/kernel/asm-offsets.c:19:
>>> /home/mchehab/redhat/trees/work/new_edac/arch/tile/include/asm/linkage.h:18:22: error: feedback.h: No such file or directory
>>> make[1]: *** [arch/tile/kernel/asm-offsets.s] Error 1
>>>
>>> Without it, it is not possible to compile anything on this archictecture
>>> using this default configuration:
>>> 	./arch/tile/configs/tilegx_defconfig
>>>
>>> Cc: stable@kernel.org
>>> Signed-of-by: Mauro Carvalho Chehab <mchehab@redhat.com>
>> Thank you, but nack.  The header is provided in the latest gcc tip code. 
> Hmm... the gcc cross-compiler I've downloaded from Tilera site about one month
> ago didn't have it (tilegx-unknown-linux-gnu-gcc (GCC) 4.4.6).

I've refreshed the toolchain pieces on the website, including the x86_64
cross-compiler, so please download the revised one next time you want to
try a build.  Thanks for your patience!

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com

Re: [PATCH] tile: remove a non-existing header

[Mauro Carvalho Chehab]

Em 12-03-2012 15:18, Chris Metcalf escreveu:
> (Cc's trimmed a bit.)
> 
> On 3/12/2012 11:00 AM, Mauro Carvalho Chehab wrote:
>> Em 12-03-2012 11:31, Chris Metcalf escreveu:
>>> On 3/12/2012 10:18 AM, Mauro Carvalho Chehab wrote:
>>>> While testing some API changes at the EDAC core, I've tested the EDAC
>>>> drivers compilation on all supported architectures, including Tilera.
>>>>
>>>> I noticed that Tilera upstream code has issues: it doesn't compile with
>>>> kernel 3.2:
>>>>
>>>> In file included from include/linux/linkage.h:5,
>>>>                  from include/linux/kernel.h:13,
>>>>                  from include/linux/sched.h:55,
>>>>                  from arch/tile/kernel/asm-offsets.c:19:
>>>> /home/mchehab/redhat/trees/work/new_edac/arch/tile/include/asm/linkage.h:18:22: error: feedback.h: No such file or directory
>>>> make[1]: *** [arch/tile/kernel/asm-offsets.s] Error 1
>>>>
>>>> Without it, it is not possible to compile anything on this archictecture
>>>> using this default configuration:
>>>> 	./arch/tile/configs/tilegx_defconfig
>>>>
>>>> Cc: stable@kernel.org
>>>> Signed-of-by: Mauro Carvalho Chehab <mchehab@redhat.com>
>>> Thank you, but nack.  The header is provided in the latest gcc tip code. 
>> Hmm... the gcc cross-compiler I've downloaded from Tilera site about one month
>> ago didn't have it (tilegx-unknown-linux-gnu-gcc (GCC) 4.4.6).
> 
> I've refreshed the toolchain pieces on the website, including the x86_64
> cross-compiler, so please download the revised one next time you want to
> try a build.  Thanks for your patience!

Thanks!
Mauro

Re: [PATCH] tile: remove a non-existing header

[Chris Metcalf]

On 3/12/2012 10:18 AM, Mauro Carvalho Chehab wrote:
> Btw, the ./arch/tile/configs/tilepro_defconfig doesn't compile even
> with this fix. It fails at arch/tile/include/asm/atomic_64.h because
> atomic64_t is never defined, but it is used there and on other places.

I suspect you may be trying to use tilepro_defconfig with ARCH=tilegx. 
It's only valid for ARCH=tilepro, just like tilegx_defconfig is only valid
for ARCH=tilegx.  There's some #ifdefs that are supposed to catch this (in
asm-offsets.c) but it's coming too late.  I'll move it so it's before all
the #includes so at least it will be the first error that you see!

> Also, while the edac_tile driver seems to be for both Tile GX and Tilepro
> sub-archs, it is not included on tilegx_defconfig. It would make sense
> to add it there, in order to make easier for people to test/use the
> EDAC drivers.

Fixed now in Linus' tree.

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com