Re: [PATCH] iomap: Set all uptodate bits for an Uptodate page

[Sedat Dilek <sedat.dilek@gmail.com>]

On Fri, Sep 25, 2020 at 3:46 PM Matthew Wilcox <willy@infradead.org> wrote:
>
> On Fri, Sep 25, 2020 at 03:36:01PM +0200, Sedat Dilek wrote:
> > > I have applied your diff on top of Linux v5.9-rc6+ together with
> > > "iomap: Set all uptodate bits for an Uptodate page".
> > >
> > > Run LTP tests:
> > >
> > > #1: syscalls (all)
> > > #2: syscalls/preadv203
> > > #3: syscalls/dirtyc0w
> > >
> > > With #1 I see some failures with madvise0x tests.
>
> Why do you think these failures are related to my patches?
>

Oh sorry, I was not saying it is related to your patches and I am not
familiar with all syscalls LTP tests.

You said:
> Qian reported preadv203.c could reproduce it easily on POWER and ARM.
> They have 64kB pages, so it's easier to hit.  You need to have a
> filesystem with block size < page size to hit the problem.

Here on my x86-64 Debian host I use Ext4-FS.
I can setup a new partition with a different filesystem if this helps.
Any recommendations?

How does the assertion look like in the logs?
You have an example.

Here on Debian I switched over to a newer kernel-libc and
kernel-headers - I guess it's good to recompile a recent LTP from Git
against this new stuff.
In my install-logs I have seen some packages where re-built against
new kernel-libc (capabilities etc.).

- Sedat -


- Sedat -