Re: [PATCH -mm] mm, huge page: Copy to access sub-page last when copy huge page

["Huang\, Ying" <ying.huang@intel.com>]

Hi, Michal,

Michal Hocko <mhocko@kernel.org> writes:

> On Fri 18-05-18 11:03:16, Huang, Ying wrote:
> [...]
>> The patch is a generic optimization which should benefit quite some
>> workloads, not for a specific use case.  To demonstrate the performance
>> benefit of the patch, we tested it with vm-scalability run on
>> transparent huge page.
>
> It is also adds quite some non-intuitive code. So is this worth? Does
> any _real_ workload benefits from the change?

I don't have any _real_ workload which benefits from this.  But I think
this is the right way to copy the huge page.  It should benefit many
workloads with heavy cache contention, as illustrated in the
micro-benchmark.  But the performance benefit may be small or
non-measurable for the _real_ workload.

The code does become not as intuitive as before.  But fortunately, all
non-intuitive code are in copy_user_huge_page(), which is a leaf
function with well defined interface and semantics.  And with the help
of the code comments, at least the intention of the code is clear.

Best Regards,
Huang, Ying

>>  include/linux/mm.h |  3 ++-
>>  mm/huge_memory.c   |  3 ++-
>>  mm/memory.c        | 43 +++++++++++++++++++++++++++++++++++++++----
>>  3 files changed, 43 insertions(+), 6 deletions(-)