Announcement

Collapse
No announcement yet.

Intel Graphics Driver's New Parallel Submission uAPI Landing For Linux 5.16

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

  • Intel Graphics Driver's New Parallel Submission uAPI Landing For Linux 5.16

    Phoronix: Intel Graphics Driver's New Parallel Submission uAPI Landing For Linux 5.16

    Intel's "parallel submission" user-space API for their i915 kernel graphics driver has been queued into DRM-Next today ahead of the Linux 5.16 kernel cycle...

    https://www.phoronix.com/scan.php?pa...Multi-LRC-Exec

  • #2
    Since it seems about kernel-user interaction, should it be based on io_uring?

    Comment


    • #3
      The existing bonding uAPI is completely broken with GuC submission --
      So here the U in uApi stands for "user space" but the U in GuC (graphics microcontroller) is actually the Greek alphabet µ and means "micro"...

      Comment


      • #4
        Originally posted by curfew View Post

        So here the U in uApi stands for "user space" but the U in GuC (graphics microcontroller) is actually the Greek alphabet µ and means "micro"...
        Yeah, a lot of people don't know the difference. µTorrent was the first popular app where people wrote ‘uTorrent’ all the time, followed by µBlock.

        Comment


        • #5
          Originally posted by cl333r View Post
          Since it seems about kernel-user interaction, should it be based on io_uring?
          No, that's not what io_uring does. It's a way for the userspace code to queue up normal I/O operations (i.e. disk & network I/O) for the kernel to perform.

          When using a GPU driver, you make ioctl calls specific to the driver or subsystem. I think GPU driver already have their own queuing mechanism. If you're interested in the details, the article has a link to the patch, which includes a lot of documentation.

          If the whole kernel interface were being redesigned, perhaps some of the experience, conventions, code, and datastructures from io_uring could contribute to the effort. I don't foresee that happening, although it'd be interesting if new drivers and subsystems did start to borrow things from it.

          Comment

          Working...
          X