Hacker News new | comments | show | ask | jobs | submitlogin
Ask HN: Is the Mill CPU Vulnerable to Spectre?
27 points by runeks a year ago | hide | past | web | 6 comments | favorite

Note that these problems are not inherent to out of order implementations. Dan Ingalls (who first implemented Smalltalk) said "you can cheat as long as you don't get caught". When speculation fails, all side effects are rolled back in x86 and ARM processors except reads to the cache since these were considered harmless. Writes to the cache are already rolled back so not doing so for reads just saves circuit complexity. There is no reason why this can't be added to the next chip designs, though nothing can be done about what is already out there.

As for the Mill, it has enough differences that these attacks wouldn't work on it. Just the separation between protection and translation, for example, would stop Meltdown in its tracks.

It is flawed to 'roll back' changes. Instead, those changes have to have never existed. Otherwise, there is a short window of time when the cache-visible side-effects have happened, and haven't been rolled back yet, and the attacker can see them (due to preemption, due to hyperthreading, due to running in another core etc).

(team Mill)

The Mill is not vulnerable to Spectre and Meltdown.

We have a paper that will go public shortly explaining how these attacks apply to in-order architectures that do support speculation, and how the Mill avoids them.

Watch this space :)

Will you be sticking to the timeline mentioned in this post: https://millcomputing.com/topic/meltdown-and-spectre/#post-3...

Will you be posting this paper on HN? I'm sure a large portion of the community here would be interested in it.

There has been some discussion of Meltdown and Spectre on the Mill forums:


It doesn't physically exist, so no.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact