Re: [tip:x86/vdso] x86, vdso: Instead of dummy functions, include < linux/spinlock_up.h>

[Stefani Seibold <stefani@seibold.net>]

Hi Peter,

Am Sonntag, den 16.02.2014, 20:06 -0800 schrieb H. Peter Anvin:
> On 02/16/2014 07:51 PM, tip-bot for H. Peter Anvin wrote:
> > Commit-ID:  bd9ee7fd99f127ee1306289415141d45792c97f3
> > Gitweb:     http://git.kernel.org/tip/bd9ee7fd99f127ee1306289415141d45792c97f3
> > Author:     H. Peter Anvin <hpa@linux.intel.com>
> > AuthorDate: Sun, 16 Feb 2014 19:47:01 -0800
> > Committer:  H. Peter Anvin <hpa@linux.intel.com>
> > CommitDate: Sun, 16 Feb 2014 19:47:01 -0800
> > 
> > x86, vdso: Instead of dummy functions, include <linux/spinlock_up.h>
> > 
> > The list of dummy functions is insufficient.  However, instead of
> > having a full list of dummy functions we can include
> > <linux/spinlock_up.h> which contains the (trivial) implementations
> > that we use on uniprocessor.
> > 
> > There aren't supposed to be any spinlocks at all in the VDSO, of
> > course.
> > 
> 
> That didn't work either.  I thought I was clever, but it didn't work at
> all.  Multiple build failures across numerous configurations.  This is
> turning into a total headache.
> 
> The "right" way to fix this is presumably to refactor a bunch of header
> files so that the vdso code doesn't have to include a bunch of kernel
> internal headers, but that is a lot of work.
> 

I think for the first time it will be okay to kick out the
_ASM_X86_SPINLOCK_H hack and accept the C=1 warnings.

At next step it is necessary to make the whole BUILD_VDSO32 path in
vclock_gettime.c independent from the kernel headers, only uapi/ should
be included.

The use of cycle_t must be replaced with u64.

We need a own copy of __native_read_tsc(), __iter_div_u64_rem, smp_rmb()
and cpu_relax().

For the non BUILD_VDSO32 path we must only move the #includes inside
this #ifndef BUILD_VDSO32

- Stefani