On Mon, 3 May 2021 03:03:48 +1000, Greg Price wrote:
>
>1. Does this new development rely on "powerful" instructions (as they
>exist in application object code) being broken down to simpler
>instructions? ("millicode"?)
>
I understand that the Classic Spectre even put Javascript at risk.
The fixes may have been in coarser granularity in Javascript
timing functions.
>2. Was not the IBM solution on the Z platform (when activated by the
>system's owner) to slow down the whole she-bang such that there was no
>timing difference based on the data? (That is, does not the "fix" simply
>stop the look-ahead based on predictiveness by just waiting for the real
>answer on stuff such that no illegal data access actually occurs?)
>
Ouch! Given the extraordinary cost of cache faults, that shouldn't be
practical. It might be better to abandon speculative execution.
<https://xkcd.com/1938/> ("Quantum-style" is metaphorical".)
-- gil
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
