[PATCH] Revert "arm64: Make default dma_ops to be noncoherent"

[Ming Lei]

On Tue, Dec 2, 2014 at 4:00 AM, Dann Frazier <dann.frazier at canonical.com> wrote:
> On Thu, Nov 27, 2014 at 8:43 PM, Ming Lei <ming.lei at canonical.com> wrote:
>> This reverts commit c7a4a7658d689f664050c45493d79adf053f226e.
>>
>> Conflicts:
>>         arch/arm64/mm/dma-mapping.c
>>
>> On ARM64, Utopic kernel ships APM's PCI host controller driver
>> which isn't upstreamed yet, and APM's ARM64 Soc is a coherent
>> platform. On the other side, ARM64 maintainer[1] said the
>> commit c7a4a7658 can be reverted safely if booting with ACPI
>> which can be supported by APM's own kernel release too.
>>
>> So please revert the commit c7a4a7658 to avoid breaking
>> PCI function on APM ARM64.
>
> Thanks Ming.

You are welcome, :-)

BTW, this patch is only for utopic.

>
> One question I have (that I also mentioned in the bug), is what will
> prevent us from having to carry this forward indefinitely?
> The m400 cartridge does not provide ACPI tables (afaik), so we need a

Actually the community thought ACPI should be used on server only,
and in case of server, the fact is that devices have cache-coherent DMA
all the time.  Since APM ARM64 is a DMA coherent SoC, it is correct to
apply the revert patch on utopic.

> device-tree based boot to continue to work. Reverting this for utopic
> might be the right stop-gap, but what's plan for vivid and beyond?

For DT based booting, my patch in the link should be correct, and
redhat shipped the similar patch in their internal tree too.  But the kernel
community thought handling the dma coherency(include irq, iommu, dma
offset, dma mask, ..) should be moved to drivers/pci of kernel first, instead
of arch/arm64.  That need cooperation between the two communities,
and I am not sure if the sort of stuff can be merged soon.

Thanks,
Ming Lei

>
>  -dann
>
>> BugLink: https://bugs.launchpad.net/bugs/1386490
>>
>> [1] http://marc.info/?l=linux-arm-kernel&m=141708838404470&w=2
>>
>> Cc: Dann Frazier <dann.frazier at canonical.com>
>> Signed-off-by: Ming Lei <ming.lei at canonical.com>
>> ---
>>  arch/arm64/mm/dma-mapping.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
>> index 4164c5a..acc807a 100644
>> --- a/arch/arm64/mm/dma-mapping.c
>> +++ b/arch/arm64/mm/dma-mapping.c
>> @@ -342,7 +342,7 @@ static int __init swiotlb_late_init(void)
>>         bus_register_notifier(&platform_bus_type, &platform_bus_nb);
>>         bus_register_notifier(&amba_bustype, &amba_bus_nb);
>>
>> -       dma_ops = &noncoherent_swiotlb_dma_ops;
>> +       dma_ops = &coherent_swiotlb_dma_ops;
>>
>>         return swiotlb_late_init_with_default_size(swiotlb_size);
>>  }
>> --
>> 1.9.1
>>
>>
>> --
>> kernel-team mailing list
>> kernel-team at lists.ubuntu.com
>> https://lists.ubuntu.com/mailman/listinfo/kernel-team