Re: [PATCH v2 26/44] mm/damon: Stop using vma_mas_store() for maple tree store

[Liam Howlett]

* SeongJae Park <sj@xxxxxxxxxx> [230105 14:33]:
> Hi Liam,
> 
> On Thu, 5 Jan 2023 19:16:00 +0000 Liam Howlett <liam.howlett@xxxxxxxxxx> wrote:
> 
> > From: "Liam R. Howlett" <Liam.Howlett@xxxxxxxxxx>
> > 
> > Prepare for the removal of the vma_mas_store() function by open coding
> > the maple tree store in this test code.
> 
> But seems this series is not really removing 'vma_mas_store()'.  Wouldn't it
> better to do the preparation and removal together in a same patch series?

It does from the all code but the nommu side.  The definition is dropped
from the header and c file in "mmap: Convert __vma_adjust() to use vma
iterator" [1].

> 
> > Set the range of the maple
> > state and call the store function directly.
> > 
> > Cc: SeongJae Park <sj@xxxxxxxxxx>
> > Cc: damon@xxxxxxxxxxxxxxx
> > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > Signed-off-by: Liam R. Howlett <Liam.Howlett@xxxxxxxxxx>
> > ---
> >  mm/damon/vaddr-test.h | 6 ++++--
> >  1 file changed, 4 insertions(+), 2 deletions(-)
> > 
> > diff --git a/mm/damon/vaddr-test.h b/mm/damon/vaddr-test.h
> > index bce37c487540..41532f7355d0 100644
> > --- a/mm/damon/vaddr-test.h
> > +++ b/mm/damon/vaddr-test.h
> > @@ -24,8 +24,10 @@ static void __link_vmas(struct maple_tree *mt, struct vm_area_struct *vmas,
> >  		return;
> >  
> >  	mas_lock(&mas);
> > -	for (i = 0; i < nr_vmas; i++)
> > -		vma_mas_store(&vmas[i], &mas);
> > +	for (i = 0; i < nr_vmas; i++) {
> > +		mas_set_range(&mas, vmas[i].vm_start, vmas[i].vm_end - 1);
> > +		mas_store_gfp(&mas, &vmas[i], GFP_KERNEL);
> > +	}
> 
> On the latest mm-unstable, vma_mas_store() uses mas_store_prealloc() instead of
> mas_store_gfp().  Seems the difference would make no problem to this test code
> in most cases, but could I ask the reason for this change?

mas_store_prealloc() expects the maple state to have the necessary
memory to store the value.  Using this function is the right way of
storing the range.  In fact, we would only need a single node since
these values will be append operations anyways.

> 
> Also, should we check the return value of mas_store_gfp()?

I can add this.  The only reason we would return an error is on ENOMEM
which seems unlikely here.  Again, it is a single node that will be
used.  The size is 256B, but it's safer to add the check.

[1] https://lore.kernel.org/linux-mm/20230105191517.3099082-28-Liam.Howlett@xxxxxxxxxx/


Thanks,
Liam