Re: [PATCH v10] mm: compaction: handle incorrect MIGRATE_UNMOVABLEtype pageblocks

[Bartlomiej Zolnierkiewicz]

On Monday 11 June 2012 03:26:49 Minchan Kim wrote:
> Hi Bartlomiej,
> 
> On 06/08/2012 05:46 PM, Bartlomiej Zolnierkiewicz wrote:
> 
> > 
> > Hi,
> > 
> > This version is much simpler as it just uses __count_immobile_pages()
> > instead of using its own open coded version and it integrates changes
> 
> 
> That's a good idea. I don't have noticed that function is there.
> When I look at the function, it has a problem, too.
> Please, look at this.
> 
> https://lkml.org/lkml/2012/6/10/180
> 
> If reviewer is okay that patch, I would like to resend your patch based on that. 

Ok, I would later merge all changes into v11 and rebase on top of your patch.

> > from Minchan Kim (without page_count change as it doesn't seem correct
> 
> 
> Why do you think so?
> If it isn't correct, how can you prevent racing with THP page freeing?

After seeing the explanation for the previous fix it is all clear now.

> > and __count_immobile_pages() does the check in the standard way; if it
> > still is a problem I think that removing 1st phase check altogether
> > would be better instead of adding more locking complexity).
> > 
> > The patch also adds compact_rescued_unmovable_blocks vmevent to vmstats
> > to make it possible to easily check if the code is working in practice.
> 
> 
> I think that part should be another patch.
> 
> 1. Adding new vmstat would be arguable so it might interrupt this patch merging.

Why would it be arguable?  It seems non-intrusive and obvious to me.

> 2. New vmstat adding is just for this patch is effective or not in real practice
>    so if we prove it in future, let's revert the vmstat. Separating it would make it
>    easily.

I would like to add this vmstat permanently, not only for the testing period..

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung Poland R&D Center
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/