Announcement

Collapse
No announcement yet.

Linux Kernel Patched For Branch History Injection "BHI" Intel CPU Vulnerability

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #21
    Originally posted by SilverBird775 View Post
    Because... this mess is unfixable? Once you give up on speed hacks the CPU will not be as fast as before. Cheating is not a flaw, it is a strategy which always fire back, indirectly. All these Vulnerabilities are not errors and have never been, it's the cheats uncovered. Both Intel and AMD has become cheat addicts and proficient at cheating but cheating in essence is not fixable, the "solutions" just stack up more cheat on cheats and on and on.

    Until the approach changes there is nothing to fix.
    Well, rather than "cheats" perhaps I would call them "shortcuts" to increase the computational power. It's a tradeoff, as it has always been, and when security is involved, there is always a minor or major performance impact.

    The "approach change" is indeed very difficult to achieve, unless you may afford to lose a lot of performance for the sake of security. You may, for example, use processors without speculative execution, without out-of-order execution, and so on... to avoid a class of vulnerablities and side channel attacks, but we all know what that means in terms of performance.

    Comment


    • #22
      Originally posted by Weasel View Post
      Even better, if you build your own kernel:

      CONFIG_SPECULATION_MITIGATIONS=n

      Turn it off via command line still adds some overhead unfortunately. To truly get rid of it you need to build it with that.
      Care to share some numbers or benchmarks? Like kernel compile vs terminal vs enabled. I would be rather curious about it.

      Comment


      • #23
        Originally posted by Weasel View Post
        Even better, if you build your own kernel:

        CONFIG_SPECULATION_MITIGATIONS=n

        Turn it off via command line still adds some overhead unfortunately. To truly get rid of it you need to build it with that.
        I don't notice any overhead. Care to elaborate?

        Comment


        • #24
          Originally posted by hf_139 View Post

          Yes, i am fine with that one.
          Since i am not living in China, it doesn't affect me at all whether or not Xi Jinping knows that i am listening to KPOP.
          Meanwhile if Western glowies know that i said a no-no-word once, they can destroy my life, lock my bank accounts, force my employer to fire me and even throw me into jail for multiple years before getting a trial.
          The Chinese glowies can tell the western glowies you said the no-no-word, because they know that and something else that the western glowies wouldn't care about.

          Also see various news articles about Chinese secret police outposts outside China that have (so far) only been used for harrassing expats.

          Comment


          • #25
            To address this question, we developed InSpectre Gadget
            Originally posted by pWe00Iri3e7Z9lHOX2Qx View Post

            Okay, that name is gold.


            Comment


            • #26
              Originally posted by blackshard View Post

              Well, rather than "cheats" perhaps I would call them "shortcuts" to increase the computational power. It's a tradeoff, as it has always been, and when security is involved, there is always a minor or major performance impact.
              Yeah they are not even shortcuts...

              It's like looking at convolution and say that using a Fourier transform is cheating. Or looking at a Fourier transform and say that the Fast Fourier transform is cheating/a shortcut. Or, for the less mathematically-inclined, being tasked to find a "value" given a str key but considering that anything else than a linear search in an array of (key, value) pairs is "a shortcut" (so Dicts and Hashmaps would be just dirty cheats).

              It's not the case. I know it's sounds like a rebuke of what you said, but on the contrary, I think you are right but a bit too meek in your assessment.

              Side effects are unfortunate and I'm as frustrated as anybody else who cares that CPU makers still don't seem to take those into account enough in their design process but assuming that anything but the naive way of doing things is cheating plainly doesn't make sense (starting with what is considered "the naive way", which is ambiguous). With that kind of view, everything is cheating/a shortcut, without exception.

              Comment


              • #27
                Originally posted by Vistaus View Post
                I don't notice any overhead. Care to elaborate?
                Some things are automatically baked in by compiler and can't be turned off. Whether you notice overhead or not is a different matter and depends on your workload too. It does have extra stuff to execute, but maybe it's insignificant for you.

                Comment


                • #28
                  Originally posted by pWe00Iri3e7Z9lHOX2Qx View Post

                  Okay, that name is gold.
                  Let's just hope there isn't actually some other tool behind the scenes doing all the work.

                  Comment

                  Working...
                  X