Announcement

Collapse
No announcement yet.

AMD EPYC 7F72 Performance On A Linux FSGSBASE-Patched Kernel

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

  • AMD EPYC 7F72 Performance On A Linux FSGSBASE-Patched Kernel

    Phoronix: AMD EPYC 7F72 Performance On A Linux FSGSBASE-Patched Kernel

    Slated for Linux 5.9 is finally mainlining the FSGSBASE patches that have been floating around the kernel mailing list for years. Testing last week showed the tentative x86/fsgsbase patches helping Intel Xeon Linux performance but with AMD also supporting this instruction set extension going back to Bulldozer, how is it looking on the likes of AMD? Here are some benchmarks.

    http://www.phoronix.com/vr.php?view=29338

  • #2
    Then again with AMD EPYC Rome already outperforming Xeon Cascade Lake (Refresh) in most workloads, any improvements like these FSGSBASE patches are just icing on the cake. More FSGSBASE tests once seeing these patches finally hit the mainline tree.
    Without full mitigations such tests are meaningless. I'm still waiting for LVI mitigation on Intel side. Otherwise it's just fooling the customers.

    Comment


    • #3
      Originally posted by Volta View Post

      Without full mitigations such tests are meaningless. I'm still waiting for LVI mitigation on Intel side. Otherwise it's just fooling the customers.
      LVI has nothing to do with FSGSBASE. Compiler flags and others at their defaults. So far no upstream software projects have switched over to using the LVI opt-in flags. As shown in various Phoronix articles, we do show what the potential for LVI mitigation to be, but it's simply not used widely yet / deemed warranted by software vendors.
      Michael Larabel
      http://www.michaellarabel.com/

      Comment


      • #4
        Originally posted by Michael View Post

        LVI has nothing to do with FSGSBASE.
        It has something common: the first one kills performance while the second one tries to catch up. It will be interesting to see the impact of LVI with FSGSBASE.

        Comment


        • #5
          How likely is a backport of this patch once it is mainlined?

          edit.: well 5.9 could be a lts kernel if we consider the +0.5 release scheme. Otherwise this fsgsbase will be interesting for webservers of any kind (nas included). users will be more in the conservative department shipping it with a lts release would be great.
          Last edited by CochainComplex; 06-29-2020, 01:37 PM.

          Comment


          • #6
            Originally posted by CochainComplex View Post
            How likely is a backport of this patch once it is mainlined?
            Probably unlikely for upstream stable kernels at least in the near term as there have been concerns expressed over possible breakage/regressions (granted I nor seemingly others commenting on these latest patches seem to be hitting issues).
            Michael Larabel
            http://www.michaellarabel.com/

            Comment


            • #7
              Originally posted by Michael View Post

              Probably unlikely for upstream stable kernels at least in the near term as there have been concerns expressed over possible breakage/regressions (granted I nor seemingly others commenting on these latest patches seem to be hitting issues).
              Ok...this makes it unlikely - So I'm hoping that 5.9 becomes a lts kernel.

              Comment

              Working...
              X