Re: [RFC PATCH] mm, proc: report PR_SET_THP_DISABLE in proc

[Michal Hocko]

On Tue 09-10-18 10:33:26, Michal Hocko wrote:
> On Thu 04-10-18 11:34:11, David Rientjes wrote:
> > On Thu, 4 Oct 2018, Michal Hocko wrote:
> > 
> > > > And prior to the offending commit, there were three ways to control thp 
> > > > but two ways to determine if a mapping was eligible for thp based on the 
> > > > implementation detail of one of those ways.
> > > 
> > > Yes, it is really unfortunate that we have ever allowed to leak such an
> > > internal stuff like VMA flags to userspace.
> > > 
> > 
> > Right, I don't like userspace dependencies on VmFlags in smaps myself, but 
> > it's the only way we have available that shows whether a single mapping is 
> > eligible to be backed by thp :/
> 
> Which is not the case due to reasons mentioned earlier. It only speaks
> about madvise status on the VMA.
> 
> > > > If there are three ways to 
> > > > control thp, userspace is still in the dark wrt which takes precedence 
> > > > over the other: we have PR_SET_THP_DISABLE but globally sysfs has it set 
> > > > to "always", or we have MADV_HUGEPAGE set per smaps but PR_SET_THP_DISABLE 
> > > > shown in /proc/pid/status, etc.
> > > > 
> > > > Which one is the ultimate authority?
> > > 
> > > Isn't our documentation good enough? If not then we should document it
> > > properly.
> > > 
> > 
> > No, because the offending commit actually changed the precedence itself: 
> > PR_SET_THP_DISABLE used to be honored for future mappings and the commit 
> > changed that for all current mappings.
> 
> Which is the actual and the full point of the fix as described in the
> changelog. The original implementation was poor and inconsistent.
> 
> > So as a result of the commit 
> > itself we would have had to change the documentation and userspace can't 
> > be expected to keep up with yet a fourth variable: kernel version.  It 
> > really needs to be simpler, just a per-mapping specifier.
> 
> As I've said, if you really need a per-vma granularity then make it a
> dedicated line in the output with a clear semantic. Do not make VMA
> flags even more confusing.

Can we settle with something please?
-- 
Michal Hocko
SUSE Labs