No announcement yet.

It Needs A Restart: GLIBC Drops Support For Restartable Sequences

  • Filter
  • Time
  • Show
Clear All
new posts

  • It Needs A Restart: GLIBC Drops Support For Restartable Sequences

    Phoronix: It Needs A Restart: GLIBC Drops Support For Restartable Sequences

    Well this is a huge Friday morning bummer: the GNU C Library (glibc) is dropping support for the very interesting Restartable Sequences (RSEQ) as some design changes need to be made, thus restarting work on restartable sequences...

  • #2
    I have just this kind of decision to make in a project at work. I would do what the GLibC people did here if left to myself. But "business" says otherwise, so we gotta deal with all the ridiculous tech debt. Ah well.


    • #3
      its good to move backward


      • #4
        Well, sometimes going back to the drawing board saves a lot of headaches in the future. I hope they will get the problems ironed out soon though.


        • #5
          Restartable Sequences were introduced into the kernel back in Linux 4.18 for allowing various performance benefits with this system call allowing for faster user-space operations --
          For work like querying the current CPU number, incrementing per-CPU counters, writing/reading to the per-CPU ring buffers, and other work --
          The performance improvements are looking very good with RSEQ in the kernel when taken advantage of.
          For making it easier to utilize Restartable Sequences from user-space, glibc was adding support for it to make it easier for developers to make use of.
          We'll see how long it takes to get the code back in there but given that it will likely be tied to a newer version of the kernel, unfortunate that it will take yet more time before widespread adoption of this feature beneficial to performance.
          What -- the -- actual -- fuck. 🤯


          • #6
            Originally posted by curfew View Post
            What -- the -- actual -- fuck. 🤯
            He let the cat run on the keyboard again.


            • #7
              I don't get it, doesn't the kernel guarantee a stable ABI? If anybody has a link to a relevant LKML discussion about possible rseq ABI changes it would be much appreciated...

              As one point of reference, the latest Docker release adds rseq to their system call whitelist, and it's in systemd's default whitelist as well. This seems to indicate that it's seeing at least some use...