Announcement

Collapse
No announcement yet.

FUTEX2 NUMA & Small Futexes Revived For Linux

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

  • FUTEX2 NUMA & Small Futexes Revived For Linux

    Phoronix: FUTEX2 NUMA & Small Futexes Revived For Linux

    Back in 2020~2021 there was lots of talk and work around FUTEX2 for improving the Linux kernel's Futex implementation for fast user mutex. The FUTEX2 work was driven in large part for helping Steam Play / Wine gaming by better matching the behavior of Microsoft Windows with its WaitForMultipleObjects handling. While the initial code landed back in Linux 5.16, there's been other remaining FUTEX2 features still desired like variable-sized futexes and NUMA-awareness. Finally now we're seeing that work revived...

    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
    almost forgot about it. Good to see that it is progressing even if slowly.

    Comment


    • #3
      > By popular demand
      > I'm not sure how much demand there actually is
      Heh, nice one. But I suppose for "helping Steam Play / Wine gaming" it would be better to finish NTSYNC.

      Comment


      • #4
        What is the status/plan for NTSYNC implementation?

        Comment


        • #5
          Originally posted by Grinness View Post
          What is the status/plan for NTSYNC implementation?
          Review-stalled I guess.

          The classic open source issue when you aren't buddies with the reviewers.

          Comment


          • #6
            Originally posted by Grinness View Post
            What is the status/plan for NTSYNC implementation?
            Last ping was in August: https://lore.kernel.org/lkml/5805970...xLoT@camazotz/

            Comment


            • #7
              Is this really needed, since it was a substandard solution to the prime issue listed: wine. To fix it properly NTSYNC was needed.

              And since NTSYNC was needed, it really provides some new functionality to the kernel. Not just shuffling around the old like FUTEX2 does.

              Comment


              • #8
                Originally posted by varikonniemi View Post
                Is this really needed, since it was a substandard solution to the prime issue listed: wine. To fix it properly NTSYNC was needed.

                And since NTSYNC was needed, it really provides some new functionality to the kernel. Not just shuffling around the old like FUTEX2 does.
                I don't think this stuff matters to WINE at all, they already got what they needed. These were side issues they tried to bring up as additional arguments about why FUTEX1 wasn't good enough for everyone, but they didn't care about themselves.

                Comment

                Working...
                X