HANG at bootup on 3.10 kernel at memset in early_alloc_aligned with CMA

HANG at bootup on 3.10 kernel at memset in early_alloc_aligned with CMA

[VIKAS SAJJAN C]

Hi,

 we are recently migrating to 3.10 kernel on a 32bit SoC, I see that while boot up, the kernel HANGS in function early_alloc_aligned() at  memset(ptr, 0, sz);

we are reserving memory using CMA 


we have 4 banks :

bank 1 : 0x200000000,  size 0x10000000

bank 2 : 0x300000000,  size 0x10000000

bank 3 : 0x400000000,  size 0x10000000

bank 4 : 0x500000000,  size 0x0f500000

 
arm_lowmem_limit is set to 0x4f000000 and mapping is created for section: 20000000 ~ 4f000000

 I came across your patch "ARM: 8356/1: mm: handle non-pmd-aligned end of RAM"

Was wondering whether patch is back ported to 3.10 stable branch, if so can you point me to the patch? if not, can you let us know the changes needed for 3.10  kernel.

HANG at bootup on 3.10 kernel at memset in early_alloc_aligned with CMA

[Mark Rutland]

On Mon, Oct 12, 2015 at 04:38:15PM +0000, VIKAS SAJJAN C wrote:
> Hi,

Hi,

>  we are recently migrating to 3.10 kernel on a 32bit SoC, I see that
>  while boot up, the kernel HANGS in function early_alloc_aligned() at
>  memset(ptr, 0, sz);
> 
> we are reserving memory using CMA 
> 
> 
> we have 4 banks :
> 
> bank 1 : 0x200000000,  size 0x10000000
> 
> bank 2 : 0x300000000,  size 0x10000000
> 
> bank 3 : 0x400000000,  size 0x10000000
> 
> bank 4 : 0x500000000,  size 0x0f500000

As this isn't 2M-aligned, this certainly could be the same issue.

> arm_lowmem_limit is set to 0x4f000000 and mapping is created for
> section: 20000000 ~ 4f000000
> 
>  I came across your patch "ARM: 8356/1: mm: handle non-pmd-aligned end
>  of RAM"
> 
> Was wondering whether patch is back ported to 3.10 stable branch, if
> so can you point me to the patch?

To the best of my knowledge, it was not. If you haven't seen it in the
v3.10.x commit logs, it presumably wasn't.

> if not, can you let us know the changes needed for 3.10  kernel.

I do not know. Have you tried backporting the patch?

I believe that you will also need to backport commit 3de1f52a3ae82326
("ARM: 8394/1: update memblock limit after mapping lowmem"). See [1,2].

Thanks,
Mark.

[1] http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/353496.html
[2] http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/353602.html