Re: [PATCH 4/4] mm,hugetlb: compute page_size_log properly

[Michal Hocko]

Sorry for a late reply, I wasn't online last week

On Fri 10-02-17 08:51:11, Davidlohr Bueso wrote:
> On Fri, 10 Feb 2017, Michal Hocko wrote:
> 
> > On Thu 09-02-17 12:53:02, Davidlohr Bueso wrote:
> > > The SHM_HUGE_* stuff  was introduced in:
> > > 
> > >    42d7395feb5 (mm: support more pagesizes for MAP_HUGETLB/SHM_HUGETLB)
> > > 
> > > It unnecessarily adds another layer, specific to sysv shm, without
> > > anything special about it: the macros are identical to the MAP_HUGE_*
> > > stuff, which in turn does correctly describe the hugepage subsystem.
> > > 
> > > One example of the problems with extra layers what this patch fixes:
> > > mmap_pgoff() should never be using SHM_HUGE_* logic. It is obviously
> > > harmless but it would still be grand to get rid of it -- although
> > > now in the manpages I don't see that happening.
> > 
> > Can we just drop SHM_HUGE_MASK altogether? It is not exported in uapi
> > headers AFAICS.
> 
> Yeah that was my original idea, however I noticed that shmget.2 mentions
> kernel internals as part of SHM_HUGE_{2MB,1GB}, ie: SHM_HUGE_SHIFT. So
> dropping _MASK doesn't make sense if we are going to keep _SHIFT.

I am not sure I understand.
$ git grep SHM_HUGE_ include/uapi/
$

So there doesn't seem to be any user visible constant. The man page
mentiones is but I do not really see how is the userspace supposed to
use it.
-- 
Michal Hocko
SUSE Labs