Re: [PATCH] ext4: avoid full buffer walks for large folio partial writes

[Jia Zhu]

On Fri, Jun 05, 2026 at 03:24:20PM +0100, Matthew Wilcox wrote:
> > The reason I am still looking at this path is that the regression is
> > visible in our LTS upgrade testing from 6.12 to 6.18.  It was introduced
> > by the ext4 large-folio enablement in v6.16.  For example, in our
> > libMicro release benchmark with THP always enabled, usecs/call, lower is
> > better:
> > 
> > case        v6.12        v6.18        regression
> > write_u1k   0.609        4.659        +665.0%
> > write_u10k  1.408        4.869        +245.8%
> 
> Ouch ;-)  No wonder you want to address this.  Do you recover all the
> regression with this fix?

With the full v2 series applied to v6.18, the small overwrite cases look
like this.  Results are usecs/call, lower is better:

case           v6.12    v6.18   v6.18 + series
write_u1k      0.609    4.659       0.528
write_u10k     1.408    4.869       0.809
pwrite_u1k     0.609    4.659       0.538
pwrite_u10k    1.399    4.889       0.819
writev_u1k     2.238    5.277       1.179
writev_u10k   11.057    8.029       4.219

This matches the regression I was trying to address.

> > The iomap conversion is the long-term fix, but it does not help kernels
> > which still use the buffer-head buffered write path.  I would like to keep
> > this as a small regression fix for that path, and make it minimal enough
> > to be suitable for stable/LTS backport.
> 
> Is it that you're using some ext4 features that aren't supported by
> iomap yet?  Could you say which ones?  That might motivate someone to
> prioritise that support.

No, this benchmark is not using a specific ext4 feature that prevents
iomap.  It is just the default ext4 buffered write path on a regular
file.

I agree that iomap looks like the better long-term direction for ext4
buffered writes.  This small fix is mainly motivated by current/LTS
kernels that still have the buffer-head path (from v6.16 through current
mainline, until ext4 buffered writes are converted to iomap upstream),
where the large-folio enablement made this tail-walk cost visible.