Announcement

Collapse
No announcement yet.

DragonFlyBSD's Meltdown Fix Causing More Slowdowns Than Linux

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

  • DragonFlyBSD's Meltdown Fix Causing More Slowdowns Than Linux

    Phoronix: DragonFlyBSD's Meltdown Fix Causing More Slowdowns Than Linux

    Following the move by Linux to introduced Kernel Page Table Isolation (KPTI) to address the Meltdown vulnerability affecting Intel CPUs, DragonFlyBSD has implemented better user/kernel separation to address this issue. While the Linux performance hit overall was minor, in our tests carried out so far the DragonFlyBSD kernel changes are causing more widespread slowdowns.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    The current workaround probably doesn't make use of PCID, which alleviates some of the performance loss, at least on the CPU's that support it.

    But not all x86 processors support it, so probably they went for the solution that's both simpler to implement and universally supported. For now.

    Wasn't KAISER/KPTI worked on for Linux since at least summer?

    From my googles, it would seem that PCID (more generically known as ASID) can be performant (or maybe less un-performant) on other architectures. Assuming that's true, hopefully this will motivate Intel to optimize PCID in addition to a proper fix for meltdown.
    ​​​
    Last edited by hiryu; 07 January 2018, 06:36 PM.

    Comment


    • #3
      Personally I can't hold any OS vendor responsible for any performance loss associated with trying to mitigate bugs in the underlying hardware. The fault needs to be laid squarely at the feet of the various hardware vendors, namely Intel, AMD, because their cpu's are also susceptible to aspects of these attack vectors, which makes sense considering they produce processors using technology licensed from Intel, ARM but not the various ARM hardware vendors because they just license the ARM technology but didn't actually develop it and NVIDIA, because the have disclosed that even their gpu's have this bug.

      But this leads me to a much bigger, over arching question, if this is really the work of a certain shadowy, massive, state actor, namely the United States government.

      Think about it for a second, every single cpu made in the last 25 years, from any vendor, including vendors that only make gpu's, have bugs that can be exploited, unbeknownst to the OS, to execute random code that can't be detected by conventional means, does this make any sense to anyone?

      We know that there are embargoes with sharing sensitive technology with certain regimes, hell you couldn't even export Playstation 1's to certain countries, maybe these bugs are the result of a covert directive to include back doors in hardware and software to aid in hacking enemy networks; maybe this was never supposed to come to light but once the cat was out of the bag there was nothing that could be done.

      Come to think of it, NVIDIA's products are probably vulnerable because they supposedly include a custom ARM based cpu integrated into the gpu.

      Comment


      • #4
        Or, you know, this stuff got really complicated over time and the time to market went down and nobody was seeing the consequences this technology introduced.

        You should also make a distinction between the bug in the Intel hardware and the bug in the AMD and ARM hardware. The Intel one is the serious one with an existing exploit and this is the one causing the huge slowdowns.

        I really wonder how you think a government could even do this 25 years long with so many people involved and everybody was OK with this and kept his mouth shut. There's no way to keep a secret with that many people involved over that period of time.

        Comment


        • #5
          Who cares Dragonfly is a hobby OS that’s unused in the enterprise.

          Comment


          • #6
            Originally posted by garegin View Post
            Who cares Dragonfly is a hobby OS that’s unused in the enterprise.
            At least it's an "OS", not just a kernel..

            I'd like to see if and how much fixing these exploits in hardware would lessen the performance gap between Intel and AMD processors. If AMD is less affected and Intel more, there should theoretically be something like close-to-parity now between newer Intel and "Zen's".

            Comment


            • #7
              The real slowdown comes after applying the Intel KPTI OS patch and the new Intel Microcode that goes along with it which makes the branch predictor in Intel CPU's significantly less aggressive. Techspot shows that the combination of the two has a more significant impact on system performance (gaming included): https://www.techspot.com/article/1556-meltdown-and-spectre-cpu-performance-windows/

              Michael, the new Intel microcode should be available from Asus for their 370 MB as a BIOS update.
              Last edited by gururise; 07 January 2018, 08:58 PM.

              Comment


              • #8
                Originally posted by droste View Post
                Or, you know, this stuff got really complicated over time and the time to market went down and nobody was seeing the consequences this technology introduced.

                You should also make a distinction between the bug in the Intel hardware and the bug in the AMD and ARM hardware. The Intel one is the serious one with an existing exploit and this is the one causing the huge slowdowns.

                I really wonder how you think a government could even do this 25 years long with so many people involved and everybody was OK with this and kept his mouth shut. There's no way to keep a secret with that many people involved over that period of time.
                Seriously? You didn't hear about Snowden? All the people at NSA save one knew what they were doing and kept their mouth shut. How is this even comparable as a moral dilemma? A "sanctioned/required" backdoor in CPUs is magnitudes lower on the moral radar.

                Comment


                • #9
                  Originally posted by droste View Post
                  I really wonder how you think a government could even do this 25 years long with so many people involved and everybody was OK with this and kept his mouth shut. There's no way to keep a secret with that many people involved over that period of time.
                  You didn't put too much thought in your response, did you? Has the government revealed everything they know about the Kennedy assassination? Ever hear of FISA courts, secret warrants, national security letters, gag orders, etc?

                  All they would need to do is issue a directive to Intel, who owns the rights to the x86 ISA that says all cpu's must have a secret back door that is undocumented and have Intel codify in their design policies that all processors have to be designed with the following "features" without disclosing to their engineers why they insist on using said designs.

                  For proof look at OpenBSD:





                  The backdoors existed for a decade until the NDA expired and the author of the backdoors revealed their existence.

                  Why do you think the Chinese have decided to use Chinese developed RISC-V based processors for their new supercomputers?

                  This all stinks to high heaven of a conspiracy.

                  Comment


                  • #10
                    Originally posted by Spooktra View Post
                    This all stinks to high heaven of a conspiracy.
                    Just wait till you read about SMM that's present since the Intel 386

                    It makes sense to have the common tools of the people compromised by the powers that be. Be it computers or the Internet in general. It's all about control and we've been doing it to ourselves since the beginning of time via various means...

                    Comment

                    Working...
                    X