Re: [PATCH 01/11] tools: Add err.h with ERR_PTR PTR_ERR interface

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Mon, Aug 31, 2015 at 09:37:18AM +0200, Jiri Olsa escreveu:
> On Fri, Aug 28, 2015 at 01:21:39PM -0300, Arnaldo Carvalho de Melo wrote:
> > Em Wed, Aug 26, 2015 at 03:46:43PM +0200, Jiri Olsa escreveu:
> > > +++ b/tools/include/linux/err.h
> > > @@ -0,0 +1,28 @@
> > > +#ifndef __TOOLS_LINUX_ERR_H
> > > +#define __TOOLS_LINUX_ERR_H
> > > +
> > > +#include <linux/compiler.h>
> > > +#include <linux/types.h>
> > > +
> > > +#include <asm/errno.h>

> > You deleted the comment in the kernel sources at this point:
 
> right.. I did not want to bring too much attention :-))

:-)

Please get the explanation about why it is safe (the unused hole bits)
and merge it with the bits from the kernel comment that make sense,
well, I think we can just do a s/Kernel//g and explain why that is so.
 
> > /*
> >  * Kernel pointers have redundant information, so we can use a
> >  * scheme where we can return either an error code or a normal
> >  * pointer with the same return value.
> >  *
> >  * This should be a per-architecture thing, to allow different
> >  * error and pointer decisions.
> >  */

> > > +#define MAX_ERRNO	4095

> > Now we're dealing with user pointers, are we completely sure we can use
> > this trick here?

> it's safe for user as well, because 'error' pointers
> fall down to the unused hole:
> 
> Documentation/x86/x86_64/mm.txt:
> ffffffffffe00000 - ffffffffffffffff (=2 MB) unused hole
> 
> haven't checked for other archs, but since it's used
> within generic code, it should be ok
> 
> I'll put the comment back with additional explanation
> wrt user space in v2

Thanks!

- Arnaldo