On Mon 29-06-20 13:18:16, Matthew Wilcox wrote:
> On Mon, Jun 29, 2020 at 08:08:51AM +0300, Mike Rapoport wrote:
> > > @@ -886,8 +868,12 @@ static struct dm_buffer *__alloc_buffer_wait_no_callback(struct dm_bufio_client
> > > return NULL;
> > >
> > > if (dm_bufio_cache_size_latch != 1 && !tried_noio_alloc) {
> > > + unsigned noio_flag;
> > > +
> > > dm_bufio_unlock(c);
> > > - b = alloc_buffer(c, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
> > > + noio_flag = memalloc_noio_save();
> >
> > I've read the series twice and I'm still missing the definition of
> > memalloc_noio_save().
> >
> > And also it would be nice to have a paragraph about it in
> > Documentation/core-api/memory-allocation.rst
>
> Documentation/core-api/gfp_mask-from-fs-io.rst:``memalloc_nofs_save``, ``memalloc_nofs_restore`` respectively ``memalloc_noio_save``,
> Documentation/core-api/gfp_mask-from-fs-io.rst: :functions: memalloc_noio_save memalloc_noio_restore
> Documentation/core-api/gfp_mask-from-fs-io.rst:allows nesting so it is safe to call ``memalloc_noio_save`` or
The patch is adding memalloc_nowait* and I suspect Mike had that in
mind, which would be a fair request. Btw. we are missing memalloc_nocma*
documentation either - I was just reminded of its existence today...
--
Michal Hocko
SUSE Labs
