In message <2c3f9748-a17f-3778-9eaa-99087f33d0e3 at FreeBSD.org>, Lev Serebryakov writes:> On 05.03.2019 22:55, Shawn Webb wrote: > > >> This came over my phone's news feed. Another example that Colin Perciv> al was right when he wrote his paper on exploiting cache for fun and prof> it many years ago. > >=20 > > Weird machines are weird. > Not-weird machines are dead-slow :-(Picking a random email in this thread to reply. The problem is that there are so many of these Spectre class of exploits that we collectively roll our eyes. Yet another one is not news any more. -- Cheers, Cy Schubert <Cy.Schubert at cschubert.com> FreeBSD UNIX: <cy at FreeBSD.org> Web: http://www.FreeBSD.org The need of the many outweighs the greed of the few.
<<On Wed, 06 Mar 2019 05:32:56 -0800, Cy Schubert <Cy.Schubert at cschubert.com> said:> In message <2c3f9748-a17f-3778-9eaa-99087f33d0e3 at FreeBSD.org>, Lev > Serebryakov > writes: >> On 05.03.2019 22:55, Shawn Webb wrote: >> >> >> This came over my phone's news feed. Another example that Colin Perciv>> al was right when he wrote his paper on exploiting cache for fun and prof>> it many years ago. >> >=20 >> > Weird machines are weird. >> Not-weird machines are dead-slow :-(> Picking a random email in this thread to reply.> The problem is that there are so many of these Spectre class of > exploits that we collectively roll our eyes. Yet another one is not > news any more.And that's likely the way it's going to be, absent some major new discovery or a complete revolution in the way we program computers (which probably puts FreeBSD out of a job). I actually attended a very interesting talk by John Hennessey today in which he discussed (at a very high level) one idea for where this goes, and it's very definitely in the mode of completely different programming models combined with completely different hardware designs. One big part of this is that more compute hardware is going the way of GPUs, where the only supported interface is provided by a blob of proprietary software so the hardware vendor is much more free to change the implementation without maintaining hardware-level (or even ISA-level) compatibility. And a lot more hardware explicit fetch/store to different levels of the memory hierarchy. -GAWollman
In message <23680.33077.883594.613635 at hergotha.csail.mit.edu>, Garrett Wollman writes:> <<On Wed, 06 Mar 2019 05:32:56 -0800, Cy Schubert <Cy.Schubert at cschubert.com> > said: > > > In message <2c3f9748-a17f-3778-9eaa-99087f33d0e3 at FreeBSD.org>, Lev > > Serebryakov > > writes: > >> On 05.03.2019 22:55, Shawn Webb wrote: > >> > >> >> This came over my phone's news feed. Another example that Colin Perciv> >> al was right when he wrote his paper on exploiting cache for fun and prof> >> it many years ago. > >> >=20 > >> > Weird machines are weird. > >> Not-weird machines are dead-slow :-( > > > Picking a random email in this thread to reply. > > > The problem is that there are so many of these Spectre class of > > exploits that we collectively roll our eyes. Yet another one is not > > news any more. > > And that's likely the way it's going to be, absent some major new > discovery or a complete revolution in the way we program computers > (which probably puts FreeBSD out of a job).Probably but likely not before I or maybe you retire. But sadly, we're becoming accustomed to the brokenness of the situation. At $JOB people who would normally be alarmed by this are shrugging it off as the new normal.> > I actually attended a very interesting talk by John Hennessey today in > which he discussed (at a very high level) one idea for where this > goes, and it's very definitely in the mode of completely different > programming models combined with completely different hardware > designs. One big part of this is that more compute hardware is going > the way of GPUs, where the only supported interface is provided by a > blob of proprietary software so the hardware vendor is much more free > to change the implementation without maintaining hardware-level (or > even ISA-level) compatibility. And a lot more hardware explicit > fetch/store to different levels of the memory hierarchy.Kind of like the Cray's of old front ended by an IBM manframe or more probably like the MASPAR (a massively parallel processor front ended by a Tru64 box where I once worked). -- Cheers, Cy Schubert <Cy.Schubert at cschubert.com> FreeBSD UNIX: <cy at FreeBSD.org> Web: http://www.FreeBSD.org The need of the many outweighs the greed of the few.