On Thu, Apr 06, 2017 at 06:09:51PM +0100, Lorenzo Pieralisi wrote:
> On Thu, Apr 06, 2017 at 05:40:16PM +0100, Russell King - ARM Linux wrote:
> > On Thu, Apr 06, 2017 at 05:21:56PM +0100, Lorenzo Pieralisi wrote:
> > > Ok, so:
> > >
> > > (1) I can do asm-generic/ioremap-nopost.h, which I assume you want to
> > > contain something like
> > >
> > > static inline void __iomem *ioremap_nopost(phys_addr_t offset, size_t size)
> > > {
> > > return ioremap_nocache(offset, size);
> > > }
> > >
> > > Funny bit is that it has to be included by asm*/io.h files _after_
> > > ioremap_nocache has been #defined (that's the reason my approach was
> > > failing miserably even on arches like eg powerpc (see [1] below) that
> > > does have ioremap_nocache),
> >
> > PowerPC does have ioremap_nocache() though:
> >
> > /**
> > * ioremap - map bus memory into CPU space
> > ...
> > * * ioremap_nocache is identical to ioremap
> > extern void __iomem *ioremap(phys_addr_t address, unsigned long size);
> > #define ioremap_nocache(addr, size) ioremap((addr), (size))
> >
> > and this include file is included very early on in linux/io.h. I don't
> > see anything that conditionalises it on anything except __KERNEL__. So,
> > the report from 0-day really doesn't make any sense to me.
> >
> > Do we know how we're ending up in linux/io.h line 169 without having
> > picked up the ioremap_nocache() definition provided by PowerPC's
> > asm/io.h ?
>
> I will debug it further but I *think* it is because:
>
> eg arch/powerpc/oprofile/op_model_cell.c includes <asm/io.h>
>
> and <asm/io.h> includes <linux/io.h> before ioremap_nocache is defined
Oh, that's just very wrong. asm/foo.h should never include linux/foo.h
especially when linux/foo.h already includes asm/foo.h. I think we
need PowerPC folk to fix this.
--
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
