LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Robin Holt <holt@sgi.com>
To: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Robin Holt <holt@sgi.com>, Christoph Lameter <clameter@sgi.com>,
	akpm@linux-foundation.org, Andrea Arcangeli <andrea@qumranet.com>,
	Avi Kivity <avi@qumranet.com>, Izik Eidus <izike@qumranet.com>,
	kvm-devel@lists.sourceforge.net,
	Peter Zijlstra <a.p.zijlstra@chello.nl>,
	general@lists.openfabrics.org,
	Steve Wise <swise@opengridcomputing.com>,
	Roland Dreier <rdreier@cisco.com>,
	Kanoj Sarcar <kanojsarcar@yahoo.com>,
	steiner@sgi.com, linux-kernel@vger.kernel.org,
	linux-mm@kvack.org, daniel.blueman@quadrics.com
Subject: Re: [patch 5/6] mmu_notifier: Support for drivers with revers maps (f.e. for XPmem)
Date: Tue, 26 Feb 2008 06:29:08 -0600	[thread overview]
Message-ID: <20080226122908.GQ11391@sgi.com> (raw)
In-Reply-To: <200802261711.33213.nickpiggin@yahoo.com.au>

> > That is it.  That is all our allowed interaction with the users process.
> 
> OK, when you said something along the lines of "the MPT library has
> control of the comm buffer", then I assumed it was an area of virtual
> memory which is set up as part of initialization, rather than during
> runtime. I guess I jumped to conclusions.

There are six regions the MPT library typically makes.  The most basic
one is a fixed size.  It describes the MPT internal buffers, the stack,
the heap, the application text, and finally the entire address space.
That last region is seldom used.  MPT only has control over the first
two.

> > That doesn't seem too unreasonable, except when you compare it to how the
> > driver currently works.  Remember, this is done from a library which has
> > no insight into what the user has done to its own virtual address space.
> > As a result, each MPI_Send() would result in a system call (or we would
> > need to have a set of callouts for changes to a processes VMAs) which
> > would be a significant increase in communication overhead.
> >
> > Maybe I am missing what you intend to do, but what we need is a means of
> > tracking one processes virtual address space changes so other processes
> > can do direct memory accesses without the need for a system call on each
> > communication event.
> 
> Yeah it's tricky. BTW. what is the performance difference between
> having a system call or no?

The system call takes many microseconds and still requires the same
latency of the communication.  Without it, our latency is
usually below two microseconds.

> > > Because you don't need to swap, you don't need coherency, and you
> > > are in control of the areas, then this seems like the best choice.
> > > It would allow you to use heap, stack, file-backed, anything.
> >
> > You are missing one point here.  The MPI specifications that have
> > been out there for decades do not require the process use a library
> > for allocating the buffer.  I realize that is a horrible shortcoming,
> > but that is the world we live in.  Even if we could change that spec,
> 
> Can you change the spec? Are you working on it?

Even if we changed the spec, the old specs will continue to be
supported.  I personally am not involved.  Not sure if anybody else is
working this issue.

> > we would still need to support the existing specs.  As a result, the
> > user can change their virtual address space as they need and still expect
> > communications be cheap.
> 
> That's true. How has it been supported up to now? Are you using
> these kind of notifiers in patched kernels?

At fault time, we check to see if it is an anon or mspec vma.  We pin
the page an insert them.  The remote OS then losses synchronicity with
the owning processes page tables.  If an unmap, madvise, etc occurs the
page tables are updated without regard to our references.  Fork or exit
(fork is caught using an LD_PRELOAD library) cause the user pages to be
recalled from the remote side and put_page returns them to the kernel.
We have documented that this loss of synchronicity is due to their
action and not supported.  Essentially, we rely upon the application
being well behaved.  To this point, that has remainded true.

Thanks,
Robin

  parent reply	other threads:[~2008-02-26 12:29 UTC|newest]

Thread overview: 91+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-02-15  6:48 [patch 0/6] MMU Notifiers V7 Christoph Lameter
2008-02-15  6:49 ` [patch 1/6] mmu_notifier: Core code Christoph Lameter
2008-02-16  3:37   ` Andrew Morton
2008-02-16  8:45     ` Avi Kivity
2008-02-16  8:56       ` Andrew Morton
2008-02-16  9:21         ` Avi Kivity
2008-02-16 10:41     ` Brice Goglin
2008-02-16 10:58       ` Andrew Morton
2008-02-16 19:31         ` Christoph Lameter
2008-02-16 19:21     ` Christoph Lameter
2008-02-17  3:01       ` Andrea Arcangeli
2008-02-17 12:24         ` Robin Holt
2008-02-17  5:04     ` Doug Maxey
2008-02-18 22:33   ` Roland Dreier
2008-02-15  6:49 ` [patch 2/6] mmu_notifier: Callbacks to invalidate address ranges Christoph Lameter
2008-02-16  3:37   ` Andrew Morton
2008-02-16 19:26     ` Christoph Lameter
2008-02-19  8:54   ` Nick Piggin
2008-02-19 13:34     ` Andrea Arcangeli
2008-02-27 22:23       ` Christoph Lameter
2008-02-27 23:57         ` Andrea Arcangeli
2008-02-19 23:08   ` Nick Piggin
2008-02-20  1:00     ` Andrea Arcangeli
2008-02-20  3:00       ` Robin Holt
2008-02-20  3:11         ` Nick Piggin
2008-02-20  3:19           ` Robin Holt
2008-02-27 22:39       ` Christoph Lameter
2008-02-28  0:38         ` Andrea Arcangeli
2008-02-27 22:35     ` Christoph Lameter
2008-02-27 22:42       ` Jack Steiner
2008-02-28  0:10       ` Christoph Lameter
2008-02-28  0:11       ` Andrea Arcangeli
2008-02-28  0:14         ` Christoph Lameter
2008-02-28  0:52           ` Andrea Arcangeli
2008-02-28  1:03             ` Christoph Lameter
2008-02-28  1:10               ` Andrea Arcangeli
2008-02-28 18:43                 ` Christoph Lameter
2008-02-29  0:55                   ` Andrea Arcangeli
2008-02-29  0:59                     ` Christoph Lameter
2008-02-29 13:13                       ` Andrea Arcangeli
2008-02-29 19:55                         ` Christoph Lameter
2008-02-29 20:17                           ` Andrea Arcangeli
2008-02-29 21:03                             ` Christoph Lameter
2008-02-29 21:23                               ` Andrea Arcangeli
2008-02-29 21:29                                 ` Christoph Lameter
2008-02-29 21:34                                 ` Christoph Lameter
2008-02-29 21:48                                   ` Andrea Arcangeli
2008-02-29 22:12                                     ` Christoph Lameter
2008-02-29 22:41                                       ` Andrea Arcangeli
2008-02-28 10:53             ` Robin Holt
2008-03-03  5:11       ` Nick Piggin
2008-03-03 19:28         ` Christoph Lameter
2008-03-03 19:50           ` Nick Piggin
2008-03-04 18:58             ` Christoph Lameter
2008-03-05  0:52               ` Nick Piggin
2008-02-15  6:49 ` [patch 3/6] mmu_notifier: invalidate_page callbacks Christoph Lameter
2008-02-16  3:37   ` Andrew Morton
2008-02-16 11:07     ` Andrea Arcangeli
2008-02-16 19:22     ` Christoph Lameter
2008-02-16 19:54       ` Avi Kivity
2008-02-19  8:46       ` Nick Piggin
2008-02-19 13:30         ` Andrea Arcangeli
2008-02-18  1:51     ` Nick Piggin
2008-02-15  6:49 ` [patch 4/6] mmu_notifier: Skeleton driver for a simple mmu_notifier Christoph Lameter
2008-02-15  6:49 ` [patch 5/6] mmu_notifier: Support for drivers with revers maps (f.e. for XPmem) Christoph Lameter
2008-02-16  3:37   ` Andrew Morton
2008-02-16 19:28     ` Christoph Lameter
2008-02-19 23:55   ` Nick Piggin
2008-02-20  3:12     ` Robin Holt
2008-02-20  3:51       ` Nick Piggin
2008-02-20  9:00         ` Robin Holt
2008-02-20  9:05           ` Robin Holt
2008-02-21  4:20           ` Nick Piggin
2008-02-21 10:58             ` Robin Holt
2008-02-26  6:11               ` Nick Piggin
2008-02-26  7:21                 ` [ofa-general] " Gleb Natapov
2008-02-26  8:52                   ` Nick Piggin
2008-02-26  9:38                     ` Gleb Natapov
2008-02-26  9:52                       ` KOSAKI Motohiro
2008-02-26 12:28                     ` Robin Holt
2008-02-26 12:29                 ` Robin Holt [this message]
2008-02-27 22:43     ` Christoph Lameter
2008-02-28  0:42       ` Andrea Arcangeli
2008-02-28  1:01         ` Christoph Lameter
2008-02-15  6:49 ` [patch 6/6] mmu_rmap_notifier: Skeleton for complex driver that uses its own rmaps Christoph Lameter
2008-02-16 10:48 ` [PATCH] KVM swapping with MMU Notifiers V7 Andrea Arcangeli
2008-02-16 11:08   ` Andrew Morton
2008-02-18 12:17     ` Andrea Arcangeli
2008-02-16 11:51   ` Robin Holt
2008-02-18 12:35     ` Andrea Arcangeli
  -- strict thread matches above, loose matches on Subject: below --
2008-02-08 22:06 [patch 0/6] MMU Notifiers V6 Christoph Lameter
2008-02-08 22:06 ` [patch 5/6] mmu_notifier: Support for drivers with revers maps (f.e. for XPmem) Christoph Lameter

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20080226122908.GQ11391@sgi.com \
    --to=holt@sgi.com \
    --cc=a.p.zijlstra@chello.nl \
    --cc=akpm@linux-foundation.org \
    --cc=andrea@qumranet.com \
    --cc=avi@qumranet.com \
    --cc=clameter@sgi.com \
    --cc=daniel.blueman@quadrics.com \
    --cc=general@lists.openfabrics.org \
    --cc=izike@qumranet.com \
    --cc=kanojsarcar@yahoo.com \
    --cc=kvm-devel@lists.sourceforge.net \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=nickpiggin@yahoo.com.au \
    --cc=rdreier@cisco.com \
    --cc=steiner@sgi.com \
    --cc=swise@opengridcomputing.com \
    --subject='Re: [patch 5/6] mmu_notifier: Support for drivers with revers maps (f.e. for XPmem)' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).