is that the “disable branch prediction altogether” flag that you said existed and I claimed might revert the processor to static prediction instead?
-
-
-
No, it's "disable cache". ;-)
-
This probably kills 5 9s of performance whereas timer interrupt every 100ns to clobber BTB would only kill 4...
-
It's really bad when your unit of performance loss is "nines"...
-
lol. "This spectre fix only has five nines of high unavailability."
-
With µc that adds the "flush btb" insn, a timer interrupt that's just "ibpb;iret" might not be in the "nines" level of horribleness, but I doubt ibpb is sufficient to clobber all branch prediction, just indirect.
-
Oh wait. If the cpu has HT, can you just steal a whole HT thread to constantly clobber BTBs?
-
This should be easy to test without kernel hacks. Pin task to odd-numbered core, set max realtime priority, spam insns that clobber btbs, run spectre test.
- 1 more reply
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.