Reply to post: Re: OK, I'll bite

Here come the lawyers! Intel slapped with three Meltdown bug lawsuits

bombastic bob Silver badge
Devil

Re: OK, I'll bite

"Get security at the cost of performance by properly flushing the pipelines between task switches."

I would think this should be done within the silicon whenever you switch 'rings'. If not the OS should most definitely do this. Does the instruction pipeline (within the silicon) stop executing properly when you switch rings, like when servicing an ISR? If not, it may be part of the Meltdown problem as well, that is the CPU generating an interrupt, which is serviced AFTER part of the pipeline executes. So reading memory generates a trigger for an ISR, but other instructions execute 'out of order' before actually servicing the ISR...

I guess these are the kinds of architecture questions that need to be asked by Intel (and others), what the safest way is to do a state change within the silicon, and how to preserve (or re-start) that state without impacting anything more than re-executing a few instructions...

So I'm guessing that this would need to happen:

a) pipeline has 'tentative' register values being stored/used by out-of-order instructions, branch predictions, etc.

b) interrupt happens, including software interrupts (executing software interrupts should happen 'in order' in my opinion, but I don't know what the silicon actually does)

c) ring switch from ISR flushes all of the 'tentative' register values, as if those instructions never executed

If that's already happening, and the spectre vulnerabilities can STILL leverage reading memory across process and kernel boundaries, then I'm confused as to how it could be mitigated at ALL...

the whole idea of instruction pipelining and branch prediction was to make it such that the software "shouldn't care" whether it exists or not. THAT also removes blame from the OS, really. But that also doesn't mean that the OS devs should sit by and let it happen [so a re-architecture is in order].

But I wouldn't blame the OS makers at all. What we were told, early on, is that this would speed up the processors WITHOUT having to re-write software. THAT was "the promise" that was broken.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019