Build error in 3.18.y-stable-queue (s390)

Build error in 3.18.y-stable-queue (s390)

[Guenter Roeck]

Hi Sasha,

s390 builds in v3.18.y-stable-queue fail as follows.

./arch/s390/include/asm/mmu_context.h: In function 'init_new_context':
./arch/s390/include/asm/mmu_context.h:33:3: error: implicit declaration of function 'mm_inc_nr_pmds'

Culprit appears to be 's390/mm: four page table levels vs. fork'. It uses mm_inc_nr_pmds(),
which does not exist in v3.18.

I noticed that arm64:allmodconfig builds fail as well, but that is not a new problem and
only seen with gcc 5.2 (or at least not with gcc 4.9.1). I'll see if I can track this down.

Guenter

Re: Build error in 3.18.y-stable-queue (s390)

[Guenter Roeck]

On 03/22/2016 08:42 PM, Guenter Roeck wrote:
> Hi Sasha,
>
> s390 builds in v3.18.y-stable-queue fail as follows.
>
> ./arch/s390/include/asm/mmu_context.h: In function 'init_new_context':
> ./arch/s390/include/asm/mmu_context.h:33:3: error: implicit declaration of function 'mm_inc_nr_pmds'
>
> Culprit appears to be 's390/mm: four page table levels vs. fork'. It uses mm_inc_nr_pmds(),
> which does not exist in v3.18.
>
> I noticed that arm64:allmodconfig builds fail as well, but that is not a new problem and
> only seen with gcc 5.2 (or at least not with gcc 4.9.1). I'll see if I can track this down.
>

Fixing the arm64 problem will require patch f5e0a12ca2d9 ("arm64: psci: move psci firmware
calls out of line"). The patch doesn't apply cleanly; there is a minor conflict in the
arm64 Makefile. Let me know if you need me to send a backport.

Thanks,
Guenter

Re: Build error in 3.18.y-stable-queue (s390)

[Sasha Levin]

Thanks for the report Guenter!

On 03/23/2016 12:21 AM, Guenter Roeck wrote:
> On 03/22/2016 08:42 PM, Guenter Roeck wrote:
>> Hi Sasha,
>>
>> s390 builds in v3.18.y-stable-queue fail as follows.
>>
>> ./arch/s390/include/asm/mmu_context.h: In function 'init_new_context':
>> ./arch/s390/include/asm/mmu_context.h:33:3: error: implicit declaration of function 'mm_inc_nr_pmds'
>>
>> Culprit appears to be 's390/mm: four page table levels vs. fork'. It uses mm_inc_nr_pmds(),
>> which does not exist in v3.18.

Should be fixed, I backported "mm: account pmd page tables to the process" as well.

>> I noticed that arm64:allmodconfig builds fail as well, but that is not a new problem and
>> only seen with gcc 5.2 (or at least not with gcc 4.9.1). I'll see if I can track this down.
>>
> 
> Fixing the arm64 problem will require patch f5e0a12ca2d9 ("arm64: psci: move psci firmware
> calls out of line"). The patch doesn't apply cleanly; there is a minor conflict in the
> arm64 Makefile. Let me know if you need me to send a backport.

I've grabbed that commit.


Thanks,
Sasha

Re: Build error in 3.18.y-stable-queue (s390)

[Guenter Roeck]

On Thu, Mar 24, 2016 at 08:38:27AM -0400, Sasha Levin wrote:
> Thanks for the report Guenter!
> 
> On 03/23/2016 12:21 AM, Guenter Roeck wrote:
> > On 03/22/2016 08:42 PM, Guenter Roeck wrote:
> >> Hi Sasha,
> >>
> >> s390 builds in v3.18.y-stable-queue fail as follows.
> >>
> >> ./arch/s390/include/asm/mmu_context.h: In function 'init_new_context':
> >> ./arch/s390/include/asm/mmu_context.h:33:3: error: implicit declaration of function 'mm_inc_nr_pmds'
> >>
> >> Culprit appears to be 's390/mm: four page table levels vs. fork'. It uses mm_inc_nr_pmds(),
> >> which does not exist in v3.18.
> 
> Should be fixed, I backported "mm: account pmd page tables to the process" as well.
> 
> >> I noticed that arm64:allmodconfig builds fail as well, but that is not a new problem and
> >> only seen with gcc 5.2 (or at least not with gcc 4.9.1). I'll see if I can track this down.
> >>
> > 
> > Fixing the arm64 problem will require patch f5e0a12ca2d9 ("arm64: psci: move psci firmware
> > calls out of line"). The patch doesn't apply cleanly; there is a minor conflict in the
> > arm64 Makefile. Let me know if you need me to send a backport.
> 
> I've grabbed that commit.
> 

Now I get:

alpha, m32r, m68k, microblaze, parisc, sparc:

mm/mmap.c: In function 'exit_mmap':
mm/mmap.c:2786:2: error: 'PUD_SHIFT' undeclared

s390:

./arch/s390/include/asm/mmu_context.h: In function 'init_new_context':
./arch/s390/include/asm/mmu_context.h:30:22: error:
	'_ASCE_TYPE_REGION3' undeclared

Guenter

Re: Build error in 3.18.y-stable-queue (s390)

[Sasha Levin]

On 03/24/2016 02:01 PM, Guenter Roeck wrote:
> On Thu, Mar 24, 2016 at 08:38:27AM -0400, Sasha Levin wrote:
>> Thanks for the report Guenter!
>>
>> On 03/23/2016 12:21 AM, Guenter Roeck wrote:
>>> On 03/22/2016 08:42 PM, Guenter Roeck wrote:
>>>> Hi Sasha,
>>>>
>>>> s390 builds in v3.18.y-stable-queue fail as follows.
>>>>
>>>> ./arch/s390/include/asm/mmu_context.h: In function 'init_new_context':
>>>> ./arch/s390/include/asm/mmu_context.h:33:3: error: implicit declaration of function 'mm_inc_nr_pmds'
>>>>
>>>> Culprit appears to be 's390/mm: four page table levels vs. fork'. It uses mm_inc_nr_pmds(),
>>>> which does not exist in v3.18.
>>
>> Should be fixed, I backported "mm: account pmd page tables to the process" as well.
>>
>>>> I noticed that arm64:allmodconfig builds fail as well, but that is not a new problem and
>>>> only seen with gcc 5.2 (or at least not with gcc 4.9.1). I'll see if I can track this down.
>>>>
>>>
>>> Fixing the arm64 problem will require patch f5e0a12ca2d9 ("arm64: psci: move psci firmware
>>> calls out of line"). The patch doesn't apply cleanly; there is a minor conflict in the
>>> arm64 Makefile. Let me know if you need me to send a backport.
>>
>> I've grabbed that commit.
>>
> 
> Now I get:
> 
> alpha, m32r, m68k, microblaze, parisc, sparc:
> 
> mm/mmap.c: In function 'exit_mmap':
> mm/mmap.c:2786:2: error: 'PUD_SHIFT' undeclared
> 
> s390:
> 
> ./arch/s390/include/asm/mmu_context.h: In function 'init_new_context':
> ./arch/s390/include/asm/mmu_context.h:30:22: error:
> 	'_ASCE_TYPE_REGION3' undeclared

Should be fixed now, I hope.

I better get more than just x86/ARM for cross compiling :/


Thanks,
Sasha

Re: Build error in 3.18.y-stable-queue (s390)

[Guenter Roeck]

Hi Sasha,

On 03/26/2016 10:41 PM, Sasha Levin wrote:
...
>
> Should be fixed now, I hope.
>

Builds now all pass, but now I get runtime errors on arm and microblaze qemu tests.

WARNING: CPU: 0 PID: 33 at mm/mmap.c:2787 exit_mmap+0x1ac/0x230()

You'll need to also pick up commit b30fe6c7ced7 ("mm: fix false-positive warning
on exit due mm_nr_pmds(mm)") to fix that.

The s390 qemu test now builds, but the test itself gets a RCU stall and hangs.
Reverting "s390/mm: four page table levels vs. fork" fixes the problem.

Guenter

Re: Build error in 3.18.y-stable-queue (s390)

[Sasha Levin]

On 03/27/2016 11:35 AM, Guenter Roeck wrote:
> Hi Sasha,
> 
> On 03/26/2016 10:41 PM, Sasha Levin wrote:
> ...
>>
>> Should be fixed now, I hope.
>>
> 
> Builds now all pass, but now I get runtime errors on arm and microblaze qemu tests.
> 
> WARNING: CPU: 0 PID: 33 at mm/mmap.c:2787 exit_mmap+0x1ac/0x230()
> 
> You'll need to also pick up commit b30fe6c7ced7 ("mm: fix false-positive warning
> on exit due mm_nr_pmds(mm)") to fix that.
> 
> The s390 qemu test now builds, but the test itself gets a RCU stall and hangs.
> Reverting "s390/mm: four page table levels vs. fork" fixes the problem.

I'm just going to remove "s390/mm: four page table levels vs. fork" and all commits
I've added as a prereq and will ask the maintainers for a proper backport.


Thanks,
Sasha