Re: [PATCH V3] KSM: allow dedup all tasks memory

From: Pavel Tatashin <pasha.tatashin@soleen.com>
To: Timofey Titovets <timofey.titovets@synesis.ru>
Cc: Oleksandr Natalenko <oleksandr@natalenko.name>,
	Jann Horn <jannh@google.com>,
	linux-doc@vger.kernel.org,
	Linux Kernel <linux-kernel@vger.kernel.org>,
	linux-mm@kvack.org, Matthew Wilcox <willy@infradead.org>
Subject: Re: [PATCH V3] KSM: allow dedup all tasks memory
Date: Tue, 13 Nov 2018 22:53:34 +0000	[thread overview]
Message-ID: <20181113225334.hnz7pqoldvvg6j3w@soleen.tm1wkky2jk1uhgkn0ivaxijq1c.bx.internal.cloudapp.net> (raw)
In-Reply-To: <CAGqmi77JMyxU9L4bZHPv4Nt=tyQsEZDQcMVMRfQ7de_LjZg+-Q@mail.gmail.com>

> > > That must work, but we out of bit space in vm_flags [1].
> > > i.e. first 32 bits already defined, and other only accessible only on
> > > 64-bit machines.
> >
> > So, grow vm_flags_t to 64-bit, or enable this feature on 64-bit only.
> 
> With all due respect to you, for that type of things we need
> mm maintainer opinion.

As far as I understood, you already got directions from the maintainers
to do similar to the way THP is implemented, and THP uses two flags:

VM_HUGEPAGE VM_NOHUGEPAGE, the same as I am thinking ksm should do if we
honor MADV_UNMERGEABLE.

When VM_NOHUGEPAGE is set khugepaged ignores those VMAs.

There may be a way to add VM_UNMERGEABLE without extending the size of
vm_flags, but that would be a good start point in looking how to add a
new flag.

Again, you could simply enable this feature on 64-bit only.

Pasha