Re: [PATCH 0/2] don't use mapcount() to check large folio sharing

From: Yin, Fengwei
Date: Fri Aug 04 2023 - 03:15:17 EST

Hi Andrew,

On 8/2/2023 8:39 PM, Yin, Fengwei wrote:
> Hi Andrew,
> On 7/29/2023 1:24 AM, Andrew Morton wrote:
>> On Sat, 29 Jul 2023 00:13:54 +0800 Yin Fengwei <fengwei.yin@xxxxxxxxx> wrote:
>>> In madvise_cold_or_pageout_pte_range() and madvise_free_pte_range(),
>>> folio_mapcount() is used to check whether the folio is shared. But it's
>>> not correct as folio_mapcount() returns total mapcount of large folio.
>>> Use folio_estimated_sharers() here as the estimated number is enough.
>> What are the user-visible runtime effects of these changes?
>> (and please try to avoid using the same Subject: for different patches)
> Can you hold on these patches to mm-unstable? I think we need to wait for
> David's work on folio_maybe_mapped_shared() and redo the fix base on that.
> Thanks and sorry for the noise.
Sorry for bothering you again for this patchset.

Let me explain the situation here:
- The reason to hold on the patches to mm-unstable is that I don't want to
promote the fix in this patch (using folio_estimated_sharers()). The
correct way is waiting for folio_maybe_mapped_shared() from David.

Merging these patches motivate using folio_estimated_sharers() in other
places. So once folio_maybe_mapped_shared() is ready, we need to replace
folio_estimated_sharers() with folio_maybe_mapped_shared().

- For this specific patches, if they are suitable for stable, we may want to
merge it (special for stable branch. I assume folio_maybe_mapped_shared()
may not be back ported to stable branch).

So how do we deal with this situation? Thanks in advance.

Yin, Fengwei

> Regards
> Yin, Fengwei