Am I the only one to notice you brought up POWER/TALOS something like five times in the last week, even when the thread originally had nothing to do with it (like this one)?
I get it you're enthusiastic about an open processor getting actually used (unlike RISCV) (and must say I am too), but it's not really an option for Qubes (which is the topic of the mailing list), so long as no-one has ported Qubes to it (and unless you have a lot of money I don't see anyone deciding to port Qubes to POWER only based on your assertions). However, even with open hardware design, all problems are not solved. For once, there is no real checking of whether the product you buy actually matches the specification you received. (And the main issue with Intel ME or Meltdown/Spectre is actually that the implementation doesn't match the spec, as the spec is safe.) For instance, I recently heard of a paper at a cryptographers' conference (don't have the reference, sorry), where researchers designed a hardware implementation of AES that worked perfectly, then changed three wires, and had a hardware implementation that still worked perfectly -- until you change a bit the frequency, and then the encryption is utterly broken. Three wires at 14nm on modern systems with the 8G transistors of POWER9, good luck to spot them. Oh, and also contrarily to what you say POWER9 is not more owner-controlled than amd64, at least according to the specification (and as stated before the implementation does not necessarily match the information you are given). That said, the two big advantages of POWER9 (or RISCV) to me are that it democratizes the idea of open hardware, and that bugs in it could maybe be found more easily than if it was closed-source (even though it's doubtful Meltdown/Spectre would have been found more easily were the implementation open -- the fact that POWER9 is also vulnerable to them is an element of proof towards that). As the chip is actually not really possible to check, it doesn't help with voluntarily inserted backdoors. Just my 2¢ :) Leo On 01/11/2018 01:25 AM, taii...@gmx.com wrote: > On 01/10/2018 05:34 PM, Vít Šesták wrote: > >> Maybe absence of suitable hardware is the reason why we don't have it. > The target I imagine would be ARM servers with performance ARM CPU's > such as the ones from Gigabyte running AppliedMicro CPU's. > > Unlike the high performance POWER these ARM CPU's suck at single > threaded tasks and are not owner controlled like POWER AFAIK so I don't > think it is worth it. > The only reason to do so would be the already available xen vs no xen > ATM for POWER - but you could definitely do it and it would run qubes > satisfactory. > > And yes ARM has a kind of IOMMU, I believe it is called GIC-v3 but not > available on the average ARM stuff like a laptop or phone. > -- You received this message because you are subscribed to the Google Groups "qubes-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to qubes-users+unsubscr...@googlegroups.com. To post to this group, send email to qubes-users@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-users/5bb92339-c4a1-3229-f086-29e089b1d578%40gaspard.io. For more options, visit https://groups.google.com/d/optout.
