Re: [RFC PATCH] mm, vmalloc: use __GFP_HIGHMEM implicitly

From: Michal Hocko
Date: Tue Feb 07 2017 - 10:04:21 EST


On Tue 07-02-17 15:24:14, Vlastimil Babka wrote:
> On 02/01/2017 03:05 PM, Michal Hocko wrote:
> > From: Michal Hocko <mhocko@xxxxxxxx>
> >
> > __vmalloc* allows users to provide gfp flags for the underlying
> > allocation. This API is quite popular
> > $ git grep "=[[:space:]]__vmalloc\|return[[:space:]]*__vmalloc" | wc -l
> > 77
> >
> > the only problem is that many people are not aware that they really want
> > to give __GFP_HIGHMEM along with other flags because there is really no
> > reason to consume precious lowmemory on CONFIG_HIGHMEM systems for pages
> > which are mapped to the kernel vmalloc space. About half of users don't
> > use this flag, though. This signals that we make the API unnecessarily
> > too complex.
> >
> > This patch simply uses __GFP_HIGHMEM implicitly when allocating pages to
> > be mapped to the vmalloc space. Current users which add __GFP_HIGHMEM
> > are simplified and drop the flag.
> >
> > Signed-off-by: Michal Hocko <mhocko@xxxxxxxx>
> > ---
> > Hi,
> > this is based on top of [1]. I believe it was Al who has brought this
> > up quite some time ago (or maybe I just misremember). The explicit
> > usage of __GFP_HIGHMEM in __vmalloc* seems to be too much to ask from
> > users. I believe there is no user which doesn't want vmalloc pages be
> > in the highmem but I might be missing something. There is vmalloc_32*
> > API but that uses GFP_DMA* explicitly which overrides __GFP_HIGHMEM. So
> > all current users _should_ be safe to use __GFP_HIGHMEM unconditionally.
> > This patch should simplify things and fix many users which consume
> > lowmem for no good reason.
> >
> > I am sending this as an RFC to get some feedback, I even haven't compile
> > tested it yet.
> >
> > Any comments are welcome.
>
> The idea sounds good. What are the potential dangers? That somebody of the
> current callers without __GFP_HIGHMEM would take a physical address of the
> page and then tried to access it via direct mapping?

Yes, that wouldn't work but I do not think anybody would want to do
something like that. Another risk would be that somebody really wanted
to use vmalloc_32* but didn't use the proper API. The physically
allocated page would then be used for a device which wouldn't be able to
access it because it would be out of its addressable space.
--
Michal Hocko
SUSE Labs