Re: [RFC PATCH 2/3] powerpc/mm: Implement pmdp_establish for ppc64

["Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>]

On 07/27/2017 06:26 PM, Michal Hocko wrote:
> On Thu 27-07-17 14:07:55, Aneesh Kumar K.V wrote:
>> We can now use this to set pmd page table entries to absolute values. THP
>> need to ensure that we always update pmd PTE entries such that we never mark
>> the pmd none. pmdp_establish helps in implementing that.
>>
>> This doesn't flush the tlb. Based on the old_pmd value returned caller can
>> decide to call flush_pmd_tlb_range()
> 
> _Why_ do we need this. It doesn't really help that the newly added
> function is not used so we could check that...


We were looking at having pmdp_establish used by the core code. But i 
guess Kirill ended up using pmdp_invalidate. If we don't have 
pmdp_establish usage in core code, we can drop this. This is to help 
Kiril make progress with series at


https://lkml.kernel.org/r/20170615145224.66200-1-kirill.shutemov@linux.intel.com


Also thinking about the interface further, I guess pmdp_establish 
interface is some what confusing. So we may want to rethink this 
further. I know that i asked for pmdp_establish in earlier review of 
Kirill's patchset. But now looking back i am not sure we can clearly 
explain only semantic requirement of pmdp_establish. One thing we may 
want to clarify is whether we should retain the Reference and change bit 
from the old entry when we are doing a pmdp_establish ?

Kirill,

Considering core code is still only using pmdp_invalidate(), we may want 
to drop this interface completely ?

-aneesh