[rfc] powerpc/npu: Cleanup MMIO ATSD flushing

[Balbir Singh]

On Tue, Oct 31, 2017 at 3:03 AM, Aneesh Kumar K.V
<aneeshkumar.opensource at gmail.com> wrote:
>
>
> On 10/30/2017 06:08 PM, Balbir Singh wrote:
>>
>> +
>> +static void pnv_npu2_invalidate_helper(struct npu_context *npu_context,
>> +               struct mm_struct *mm, unsigned long start,
>> +               unsigned long end, bool flush)
>> +{
>> +       unsigned long address;
>> +       bool is_thp;
>> +       unsigned int hshift, shift;
>> +
>> +       address = start;
>> +       do {
>> +               local_irq_disable();
>> +               find_linux_pte(mm->pgd, address, &is_thp, &hshift);
>> +               if (!is_thp)
>> +                       shift = PAGE_SHIFT;
>> +               else
>> +                       shift = hshift;
>
>
> Is that correct? if is_thp is 0 can we derive shift from hshift? IIUC we set
> hshift only
> if it is a hugepage.

OK.. I assumed that it'll be set. So I've got to check for


if (is_thp)
      shift = mmu_psize_defs[MMU_PAGE_2M].shift
else if (hshift)
    shift = hshift
else
    shift = PAGE_SHIFT

>
>
>> +               mmio_invalidate(npu_context, address > 0, address, flush,
>> +                               shift);
>> +               local_irq_enable();
>> +               address += (1ull << shift);
>> +       } while (address < end);
>>   }
>>

Thanks for the review!

Balbir Singh.