Announcement

Collapse
No announcement yet.

Bisected: The Unfortunate Reason Linux 4.20 Is Running Slower

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

  • #41
    Originally posted by TemplarGR View Post

    They have lost in court for that. Compiler shenanigans and stuff...
    That was mostly specific to Windows machines. Linux systems almost always use open source compilers that did not do that.

    Comment


    • #42
      Originally posted by Michael View Post

      That's because according to Intel, the current mitigations already cover these new attack vectors.
      According to the researchers, the attacks work on hardware with full migrations. :/

      Comment


      • #43
        Originally posted by ryao View Post
        It looks like this should affect AMD hardware too:


        https://arstechnica.com/gadgets/2018...t-performance/

        My guess is is that Michael is testing AMD hardware that lacks the microcode update that gives stibp support. AMD’s website says that he needs to update his BIOS to get it:

        https://www.amd.com/en/corporate/security-updates
        Michael I think that your AMD benchmarks might not be correct. If your system has the latest AMD microcode from a bios update, it should support STIBP and this patch should turn that on.

        Comment


        • #44
          See what I mean with idiots spreading bullshit on this forum about how it only affects Intel?

          Comment


          • #45
            Originally posted by ryao View Post
            Michael I think that your AMD benchmarks might not be correct. If your system has the latest AMD microcode from a bios update, it should support STIBP and this patch should turn that on.
            All of my AMD systems were tested with their latest BIOS releases.
            Michael Larabel
            https://www.michaellarabel.com/

            Comment


            • #46
              Originally posted by Michael View Post

              All of my AMD systems were tested with their latest BIOS releases.
              Michael Is stibp missing from /proc/cpuinfo? It might be that the motherboard manufacturer has not provided an update containing the latest microcode.

              It looks like it is possible to install the amd64-microcode package on Ubuntu to get the fix. Debian suggests that just installing the package and rebooting would make it work:



              I am not sure if it is that simple (i.e. does a boot script need to be enabled?), but if dmesg shows an update was done that was not there on a prior boot and /proc/cpuinfo starts showing stibp, you should have it.
              Last edited by ryao; 16 November 2018, 06:54 PM.

              Comment


              • #47
                Originally posted by Michael View Post

                That's because according to Intel, the current mitigations already cover these new attack vectors.
                Michael Since I now know that I can highlight you and you tend to only read my comments when I highlight you (not unreasonable), I probably should state that Intel is reportedly spreading misinformation about the latest Spectre and meltdown variants. The researchers that found them claim to have verified that they work against fully patched systems that have the current mitigations enabled.

                https://www.theregister.co.uk/2018/1...down_variants/
                Last edited by ryao; 16 November 2018, 07:12 PM.

                Comment


                • #48
                  Originally posted by Weasel View Post
                  See what I mean with idiots spreading bullshit on this forum about how it only affects Intel?
                  Part of it is AMD’s initial insistence that Spectre v2 does not apply to their hardware, followed by insistence that it is rare (like it is not rare on Intel) and then discouragement of the deployment of firmware updates providing mitigations stating that they are unnecessary. I suspect that last part might have influenced Michael’s benchmark data.

                  Until things have been properly scrutinized and re-evaluated, we will not know the extent to which AMD’s performance is affected by this patch. It could also be that many systems where it should be enabled by the logic in the kernel won’t have it enabled due to AMD discouraging the propagation of the microcode updates that provide the mitigations. The impact to the general public would be unfairly lessened by that.

                  Comment


                  • #49
                    Originally posted by ryao View Post

                    Why pre-Haswell? Intel decided to to avoid developing fixes for anything older than Sandy Bridge.



                    i would think it would be worse for anything pre-Sandy Bridge.
                    You are incorrect, your source also states that updated microcode is available for Nehalem and Westmere - see also this newer file from August 2018 from Intel: https://www.intel.com/content/dam/ww...e-guidance.pdf

                    Haswell introduced the INVPCID instruction which is used to re-gain some performance for some mitigations. Hence older processors than Haswell with the mitigations enabled should lose more performance than newer ones and I'd like to see the exact numbers. What that means: The performance implications should be even worse for these older systems.

                    Comment


                    • #50
                      Someone should make unoptimized -O0 distro, so that people have got real secure un/performance right from the beginning
                      Last edited by dungeon; 16 November 2018, 08:18 PM.

                      Comment

                      Working...
                      X