On Mon, Oct 31, 2022 at 03:11:38PM +0000, David Howells wrote:
> Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx> wrote:
>
> > +static inline __must_check void *folio_map_local(struct folio *folio)
> > +{
> > + might_alloc(GFP_KERNEL);
> > +
> > + if (!IS_ENABLED(CONFIG_HIGHMEM))
> > + return folio_address(folio);
>
> Can you make the might_alloc() contingent on CONFIG_HIGHMEM=y?
... so that 64-bit builds can inadvertently introduce bugs? Either it's
safe to do GFP_KERNEL allocations from a given context, or it's not.
That's not going to depend on CONFIG_HIGHMEM.
