old MirOS News

Sponsored by
HostEurope Logo

old MirOS News

⚠ This page contains old, outdated, obsolete, … historic or WIP content! No warranties e.g. for correctness!

All 1 2 3 4 5 6 7 8 9 10 11 12 13 14

The Intelpocalypse

2018-01-04
Tags: archaeology bug grml hardware pcli security

The unveiling of the three new CPU bug classes, collected in the two brandbugs “Meltdown” and “Spectre”, has mostly shocked the BSDs; I’ve got it on some authority that even FreeBSD was not informed ahead of time, left alone the others. Thanks to laffer1 from MidnightBSD for a couple of heads-up warnings into our direction!

Here’s what I could gather until now (please do correct me if I’m wrong):

Meltdown is specific to Intel® CPUs with out-of-order execution, that is, all P6-class (Pentium Pro/MMX, Pentium Ⅱ, but not Pentium Ⅰ/MMX) or newer (except old Atom) CPUs. It appears to allow user processes to read kernel memory, but not across VMs, nor to attack a hypervisor. A variant for ARM exists but AMD’s x86 CPUs are supposedly safe. The KAISER/FUCKWIT/UASS/KPTI patches for Linux fix this, at huge performance cost on x86, not so much on ARM, and no cost for unaffected CPU models (runtime detected).

Spectre affects x86, ARM, POWER CPUs and possibly others. I’ve not yet found information on whether it is also limited to CPUs with out-of-order executions, but it seems likely. SPARC CPUs might be safe; Solaris/SPARC64 is safe due to the way its memory addressing works. If the OOO execution assumption is true, 80486 and P5 class x86 CPUs are also safe. This one does allow cross-VM and hypervisor attacks, so if the bare metal CPU is vulnerable, SOL. There does not yet seem to be a generic fix; some hint at having to patch the compiler and recompile everything with a workaround that has a performance cost, even if the CPU is not affected, or was fixed with a microcode update. AMD’s x86 CPUs are partially hit, one of the variants does not work on them.

“CERT recommends throwing away your CPU and buying an non-vulnerable one” (thanks to El Reg), but nobody states which CPUs are not vulnerable.

At the present time, we suggest any MirBSD/i386 instances that run on any CPU other than an 80486 or P5-class (Pentium Ⅰ or a non-PPro MMX) to be restricted to single user or trusted user access only, and no untrusted software including ECMAscript to be run on them.

Watch this space for updates. Oh, and, if you know what you’re (and I’m) talking about, please, again, do provide me with information necessary to provide those updates, both to MirBSD and to this space. Thank you.

MirOS Logo