Re: [f2fs-dev] [PATCH v2] f2fs: another way to set large folio by remembering inode number

[Jaegeuk Kim]

On 05/22, Matthew Wilcox wrote:
> On Fri, May 22, 2026 at 03:59:45AM +0000, Jaegeuk Kim wrote:
> > On 05/21, Matthew Wilcox wrote:
> > > On Thu, May 21, 2026 at 11:57:48AM -0400, Theodore Tso wrote:
> > > > So let me get this straight.  This is a magic xattr interface which is
> > > > not even persisted in the file system, but instead sets a 32-bit
> > > > bitmask in the struct inode which disappears once the inode gets
> > > > flushed from the inode stack.  And it uses a generic xattr name,
> > > > "user.fadvise".
> > > > 
> > > > There's no way in *hell* any other file system is likely to adopt such
> > > > a broken interface, so why didn't you just use an ioctl to set this
> > > > magic f2fs-specific flag?
> > > 
> > > I mean, yes, this API is horrendous.  But it's just another example of
> > > f2fs thinking it's somehow special and not just enabling large folios
> > > like other filesystems do.  This hurts everyone, not just people who use
> > > f2fs.
> > 
> > >From the production viewpoint, I raised a concern on setting large folio by
> > default, since that exhausts lots of high-order pages, which were needed for
> > essential system services and critical apps.
> 
> Random fears or actual data?

This was a quick buddyinfo right after booting the device.

Before:
Node 0, zone   Normal  22684  42284  28704  16901   9515   4566   1854    673    181     36    758

After disabling EROFS large folio:
Node 0, zone   Normal   8486   4732   2175   1161    697    272     82     19      3      1    856

> 
> 
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel