[PATCH] powerpc: don't re-issue spinlock typedef that breaks older gcc

[PATCH] powerpc: don't re-issue spinlock typedef that breaks older gcc

[Paul Gortmaker]

Commit b3084f4db3aeb991c507ca774337c7e7893ed04f ("powerpc/thp: Fix
crash on mremap") added a "typedef struct spinlock spinlock_t;"
which on gcc 4.5.2 (and possibly other versions) causes many of:

include/linux/spinlock_types.h:76:3: error: redefinition of typedef 'spinlock_t'
arch/powerpc/include/asm/pgtable-ppc64.h:563:25: note: previous declaration of 'spinlock_t' was here
In file included from include/linux/mutex.h:15:0,
                 from include/linux/notifier.h:13,
                 from include/linux/pm_qos.h:8,
                 from include/linux/netdevice.h:28,
                 from drivers/net/wireless/ath/wil6210/wil6210.h:20,
                 from drivers/net/wireless/ath/wil6210/debug.c:17:

It appears that somewhere between gcc 4.5.2 and 4.6.3 this
redefinition restriction was lifted.  Using the proper header
from within !ASSEMBLY seems to fix it up in an acceptable way.

Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
---

[ Note that b3084f4db3 isn't mainline yet, it is currently in
  benh/powerpc.git #merge -- but is headed there soon via:
             https://lkml.org/lkml/2014/1/27/599           ]

 arch/powerpc/include/asm/pgtable-ppc64.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/include/asm/pgtable-ppc64.h b/arch/powerpc/include/asm/pgtable-ppc64.h
index d27960c89a71..3b638411646a 100644
--- a/arch/powerpc/include/asm/pgtable-ppc64.h
+++ b/arch/powerpc/include/asm/pgtable-ppc64.h
@@ -111,6 +111,8 @@
 
 #ifndef __ASSEMBLY__
 
+#include <linux/spinlock_types.h>
+
 /*
  * This is the default implementation of various PTE accessors, it's
  * used in all cases except Book3S with 64K pages where we have a
@@ -560,7 +562,6 @@ extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 			    pmd_t *pmdp);
 
 #define pmd_move_must_withdraw pmd_move_must_withdraw
-typedef struct spinlock spinlock_t;
 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
 					 spinlock_t *old_pmd_ptl)
 {
-- 
1.8.5.2

Re: [PATCH] powerpc: don't re-issue spinlock typedef that breaks older gcc

[Aneesh Kumar K.V]

Paul Gortmaker <paul.gortmaker@windriver.com> writes:

> Commit b3084f4db3aeb991c507ca774337c7e7893ed04f ("powerpc/thp: Fix
> crash on mremap") added a "typedef struct spinlock spinlock_t;"
> which on gcc 4.5.2 (and possibly other versions) causes many of:
>
> include/linux/spinlock_types.h:76:3: error: redefinition of typedef 'spinlock_t'
> arch/powerpc/include/asm/pgtable-ppc64.h:563:25: note: previous declaration of 'spinlock_t' was here
> In file included from include/linux/mutex.h:15:0,
>                  from include/linux/notifier.h:13,
>                  from include/linux/pm_qos.h:8,
>                  from include/linux/netdevice.h:28,
>                  from drivers/net/wireless/ath/wil6210/wil6210.h:20,
>                  from drivers/net/wireless/ath/wil6210/debug.c:17:
>
> It appears that somewhere between gcc 4.5.2 and 4.6.3 this
> redefinition restriction was lifted.  Using the proper header
> from within !ASSEMBLY seems to fix it up in an acceptable way.
>
> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
> ---
>


http://mid.gmane.org/1389939036.3000.7.camel@ThinkPad-T5421.cn.ibm.com

This was posted earlier.



> [ Note that b3084f4db3 isn't mainline yet, it is currently in
>   benh/powerpc.git #merge -- but is headed there soon via:
>              https://lkml.org/lkml/2014/1/27/599           ]
>
>  arch/powerpc/include/asm/pgtable-ppc64.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/arch/powerpc/include/asm/pgtable-ppc64.h b/arch/powerpc/include/asm/pgtable-ppc64.h
> index d27960c89a71..3b638411646a 100644
> --- a/arch/powerpc/include/asm/pgtable-ppc64.h
> +++ b/arch/powerpc/include/asm/pgtable-ppc64.h
> @@ -111,6 +111,8 @@
>
>  #ifndef __ASSEMBLY__
>
> +#include <linux/spinlock_types.h>
> +
>  /*
>   * This is the default implementation of various PTE accessors, it's
>   * used in all cases except Book3S with 64K pages where we have a
> @@ -560,7 +562,6 @@ extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
>  			    pmd_t *pmdp);
>
>  #define pmd_move_must_withdraw pmd_move_must_withdraw
> -typedef struct spinlock spinlock_t;
>  static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
>  					 spinlock_t *old_pmd_ptl)
>  {
> -- 
> 1.8.5.2

Re: [PATCH] powerpc: don't re-issue spinlock typedef that breaks older gcc

[Paul Gortmaker]

On 14-01-28 12:28 PM, Aneesh Kumar K.V wrote:
> Paul Gortmaker <paul.gortmaker@windriver.com> writes:
> 
>> Commit b3084f4db3aeb991c507ca774337c7e7893ed04f ("powerpc/thp: Fix
>> crash on mremap") added a "typedef struct spinlock spinlock_t;"
>> which on gcc 4.5.2 (and possibly other versions) causes many of:
>>
>> include/linux/spinlock_types.h:76:3: error: redefinition of typedef 'spinlock_t'
>> arch/powerpc/include/asm/pgtable-ppc64.h:563:25: note: previous declaration of 'spinlock_t' was here
>> In file included from include/linux/mutex.h:15:0,
>>                  from include/linux/notifier.h:13,
>>                  from include/linux/pm_qos.h:8,
>>                  from include/linux/netdevice.h:28,
>>                  from drivers/net/wireless/ath/wil6210/wil6210.h:20,
>>                  from drivers/net/wireless/ath/wil6210/debug.c:17:
>>
>> It appears that somewhere between gcc 4.5.2 and 4.6.3 this
>> redefinition restriction was lifted.  Using the proper header
>> from within !ASSEMBLY seems to fix it up in an acceptable way.
>>
>> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
>> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
>> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
>> ---
>>
> 
> 
> http://mid.gmane.org/1389939036.3000.7.camel@ThinkPad-T5421.cn.ibm.com
> 
> This was posted earlier.

I see.  Well I guess Ben didn't use it since it is the same as the
temporary not-signed-off-by hack patch I posted earlier as well.

https://lkml.org/lkml/2014/1/27/584

I believe what I've posted here below to be the proper fix.

Paul.
--


> 
> 
> 
>> [ Note that b3084f4db3 isn't mainline yet, it is currently in
>>   benh/powerpc.git #merge -- but is headed there soon via:
>>              https://lkml.org/lkml/2014/1/27/599           ]
>>
>>  arch/powerpc/include/asm/pgtable-ppc64.h | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/include/asm/pgtable-ppc64.h b/arch/powerpc/include/asm/pgtable-ppc64.h
>> index d27960c89a71..3b638411646a 100644
>> --- a/arch/powerpc/include/asm/pgtable-ppc64.h
>> +++ b/arch/powerpc/include/asm/pgtable-ppc64.h
>> @@ -111,6 +111,8 @@
>>
>>  #ifndef __ASSEMBLY__
>>
>> +#include <linux/spinlock_types.h>
>> +
>>  /*
>>   * This is the default implementation of various PTE accessors, it's
>>   * used in all cases except Book3S with 64K pages where we have a
>> @@ -560,7 +562,6 @@ extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
>>  			    pmd_t *pmdp);
>>
>>  #define pmd_move_must_withdraw pmd_move_must_withdraw
>> -typedef struct spinlock spinlock_t;
>>  static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
>>  					 spinlock_t *old_pmd_ptl)
>>  {
>> -- 
>> 1.8.5.2
> 

Re: [PATCH] powerpc: don't re-issue spinlock typedef that breaks older gcc

[Aneesh Kumar K.V]

Paul Gortmaker <paul.gortmaker@windriver.com> writes:

> On 14-01-28 12:28 PM, Aneesh Kumar K.V wrote:
>> Paul Gortmaker <paul.gortmaker@windriver.com> writes:
>> 
>>> Commit b3084f4db3aeb991c507ca774337c7e7893ed04f ("powerpc/thp: Fix
>>> crash on mremap") added a "typedef struct spinlock spinlock_t;"
>>> which on gcc 4.5.2 (and possibly other versions) causes many of:
>>>
>>> include/linux/spinlock_types.h:76:3: error: redefinition of typedef 'spinlock_t'
>>> arch/powerpc/include/asm/pgtable-ppc64.h:563:25: note: previous declaration of 'spinlock_t' was here
>>> In file included from include/linux/mutex.h:15:0,
>>>                  from include/linux/notifier.h:13,
>>>                  from include/linux/pm_qos.h:8,
>>>                  from include/linux/netdevice.h:28,
>>>                  from drivers/net/wireless/ath/wil6210/wil6210.h:20,
>>>                  from drivers/net/wireless/ath/wil6210/debug.c:17:
>>>
>>> It appears that somewhere between gcc 4.5.2 and 4.6.3 this
>>> redefinition restriction was lifted.  Using the proper header
>>> from within !ASSEMBLY seems to fix it up in an acceptable way.
>>>
>>> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
>>> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
>>> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>>> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
>>> ---
>>>
>> 
>> 
>> http://mid.gmane.org/1389939036.3000.7.camel@ThinkPad-T5421.cn.ibm.com
>> 
>> This was posted earlier.
>
> I see.  Well I guess Ben didn't use it since it is the same as the
> temporary not-signed-off-by hack patch I posted earlier as well.
>
> https://lkml.org/lkml/2014/1/27/584
>
> I believe what I've posted here below to be the proper fix.

I had another variant which needed this

http://mid.gmane.org/1388999012-14424-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com

BTW I had added the above struct spinlock; patch as the backport to
stable 3.13 series. So if we are picking another one, we may need to
update stable also

-aneesh

Re: [PATCH] powerpc: don't re-issue spinlock typedef that breaks older gcc

[Paul Gortmaker]

On 14-01-28 01:49 PM, Aneesh Kumar K.V wrote:
> Paul Gortmaker <paul.gortmaker@windriver.com> writes:
> 
>> On 14-01-28 12:28 PM, Aneesh Kumar K.V wrote:
>>> Paul Gortmaker <paul.gortmaker@windriver.com> writes:
>>>
>>>> Commit b3084f4db3aeb991c507ca774337c7e7893ed04f ("powerpc/thp: Fix
>>>> crash on mremap") added a "typedef struct spinlock spinlock_t;"
>>>> which on gcc 4.5.2 (and possibly other versions) causes many of:
>>>>
>>>> include/linux/spinlock_types.h:76:3: error: redefinition of typedef 'spinlock_t'
>>>> arch/powerpc/include/asm/pgtable-ppc64.h:563:25: note: previous declaration of 'spinlock_t' was here
>>>> In file included from include/linux/mutex.h:15:0,
>>>>                  from include/linux/notifier.h:13,
>>>>                  from include/linux/pm_qos.h:8,
>>>>                  from include/linux/netdevice.h:28,
>>>>                  from drivers/net/wireless/ath/wil6210/wil6210.h:20,
>>>>                  from drivers/net/wireless/ath/wil6210/debug.c:17:
>>>>
>>>> It appears that somewhere between gcc 4.5.2 and 4.6.3 this
>>>> redefinition restriction was lifted.  Using the proper header
>>>> from within !ASSEMBLY seems to fix it up in an acceptable way.
>>>>
>>>> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
>>>> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
>>>> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>>>> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
>>>> ---
>>>>
>>>
>>>
>>> http://mid.gmane.org/1389939036.3000.7.camel@ThinkPad-T5421.cn.ibm.com
>>>
>>> This was posted earlier.
>>
>> I see.  Well I guess Ben didn't use it since it is the same as the
>> temporary not-signed-off-by hack patch I posted earlier as well.
>>
>> https://lkml.org/lkml/2014/1/27/584
>>
>> I believe what I've posted here below to be the proper fix.
> 
> I had another variant which needed this
> 
> http://mid.gmane.org/1388999012-14424-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com

What config did you use to trigger that?  I've not seen it in
allyes/allmodconfig.  I'd like us to try and fix it an alternate
way, vs. fragmenting the header into smaller and smaller
specialized chunks, if possible.

> 
> BTW I had added the above struct spinlock; patch as the backport to
> stable 3.13 series. So if we are picking another one, we may need to
> update stable also

The stable tree is self-correcting ; it won't take any patches that
don't have the same commit present in mainline.  But yes, someone
will still have to _nominate_ one for stable tree consideration.

Paul.
--

> 
> -aneesh
>