On Wed, 29 Oct 2025, Christoph Hellwig wrote:
> On Wed, Oct 29, 2025 at 02:31:05PM +0100, Rafael J. Wysocki wrote:
> > > This commit fixes the suspend code so that it issues flushes before
> > > writing the header and after writing the header.
> >
> > Hmm, shouldn't it flush every time it does a sync write, and not just
> > in these two cases?
>
> It certainly should not use the PREFLUSH flag that flushes before
> writing, as the cache will be dirty again after that.
>
> I'd expect a single blkdev_issue_flush after all writing is done,
> under the assumption that the swsusp swap writing doesn't have
> transaction integrity for individual writes anyway.
I think that we should use two flushes - one before writing the header and
the other after writing the header. Otherwise, it could be possible that
the header is written and some of the data is not written, if the system
loses power during hibernation.
Mikulas
