mbox series

[0/6] mm: convert numa balancing functions to use a folio

Message ID 20230918103213.4166210-1-wangkefeng.wang@huawei.com
Headers
Series mm: convert numa balancing functions to use a folio |

Message

Kefeng Wang Sept. 18, 2023, 10:32 a.m. UTC 
  The do_numa_pages only handle non-compound page, and only PMD-mapped THP
is handled in do_huge_pmd_numa_page(), but large, PTE-mapped folio will
be supported, let's convert more numa balancing functions to use/take a
folio in preparation for that, no functional change intended for now.

Kefeng Wang (6):
  sched/numa, mm: make numa migrate functions to take a folio
  mm: mempolicy: make mpol_misplaced() to take a folio
  mm: memory: make numa_migrate_prep() to take a folio
  mm: memory: use a folio in do_numa_page()
  mm: memory: add vm_normal_pmd_folio()
  mm: huge_memory: use a folio in do_huge_pmd_numa_page()

 include/linux/mempolicy.h            |  4 +--
 include/linux/mm.h                   |  2 ++
 include/linux/sched/numa_balancing.h |  4 +--
 kernel/sched/fair.c                  | 12 +++----
 mm/huge_memory.c                     | 28 ++++++++--------
 mm/internal.h                        |  2 +-
 mm/memory.c                          | 49 ++++++++++++++++------------
 mm/mempolicy.c                       | 20 ++++++------
 8 files changed, 65 insertions(+), 56 deletions(-)

Comments

Matthew Wilcox Sept. 18, 2023, 12:57 p.m. UTC | #1 
On Mon, Sep 18, 2023 at 06:32:07PM +0800, Kefeng Wang wrote:
> The do_numa_pages only handle non-compound page, and only PMD-mapped THP
> is handled in do_huge_pmd_numa_page(), but large, PTE-mapped folio will
> be supported, let's convert more numa balancing functions to use/take a
> folio in preparation for that, no functional change intended for now.
> 
> Kefeng Wang (6):
>   sched/numa, mm: make numa migrate functions to take a folio
>   mm: mempolicy: make mpol_misplaced() to take a folio
>   mm: memory: make numa_migrate_prep() to take a folio
>   mm: memory: use a folio in do_numa_page()
>   mm: memory: add vm_normal_pmd_folio()
>   mm: huge_memory: use a folio in do_huge_pmd_numa_page()

This all seems OK.  It's kind of hard to review though because you change
the same line multiple times.  I think it works out better to go top-down
instead of bottom-up.  That is, start with do_numa_page() and pass
&folio->page to numa_migrate_prep.  Then do vm_normal_pmd_folio() followed
by do_huge_pmd_numa_page().  Fourth would have been numa_migrate_prep(),
etc.  I don't want to ask you to redo the entire series, but for future
patch series.

Also, it's nce to do things like remove the unnecessary 'extern' from
function declarations when you change them from page to folio.  And
please try to stick to 80 columns; I know it's not always easy/possible.
Kefeng Wang Sept. 18, 2023, 11:59 p.m. UTC | #2 
On 2023/9/18 20:57, Matthew Wilcox wrote:
> On Mon, Sep 18, 2023 at 06:32:07PM +0800, Kefeng Wang wrote:
>> The do_numa_pages only handle non-compound page, and only PMD-mapped THP
>> is handled in do_huge_pmd_numa_page(), but large, PTE-mapped folio will
>> be supported, let's convert more numa balancing functions to use/take a
>> folio in preparation for that, no functional change intended for now.
>>
>> Kefeng Wang (6):
>>    sched/numa, mm: make numa migrate functions to take a folio
>>    mm: mempolicy: make mpol_misplaced() to take a folio
>>    mm: memory: make numa_migrate_prep() to take a folio
>>    mm: memory: use a folio in do_numa_page()
>>    mm: memory: add vm_normal_pmd_folio()
>>    mm: huge_memory: use a folio in do_huge_pmd_numa_page()
> 
> This all seems OK.  It's kind of hard to review though because you change
> the same line multiple times.  I think it works out better to go top-down
> instead of bottom-up.  That is, start with do_numa_page() and pass
> &folio->page to numa_migrate_prep.  Then do vm_normal_pmd_folio() followed
> by do_huge_pmd_numa_page().  Fourth would have been numa_migrate_prep(),
> etc.  I don't want to ask you to redo the entire series, but for future
> patch series.
> 
> Also, it's nce to do things like remove the unnecessary 'extern' from
> function declarations when you change them from page to folio.  And
> please try to stick to 80 columns; I know it's not always easy/possible.
> 

Thanks for your review and suggestion, I will keep them in mind when
sending new patch, thanks.