Announcement

Collapse
No announcement yet.

Collabora Looks To Add DMA-BUF Protocol To Wayland

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

  • Collabora Looks To Add DMA-BUF Protocol To Wayland

    Phoronix: Collabora Looks To Add DMA-BUF Protocol To Wayland

    Collabora developers are looking at adding a generic DMA-BUF protocol to Wayland, which could benefit media players and other applications...

    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
    Is this the first step towards being able to use multiple GPUs on WL or am i completelty confusing DMA_BUF with something else??

    Comment


    • #3
      Hi all,

      can someone explain why we need DMA BUF ? What problems it solves and what possible new issues it might create ? I believe it was for security reasons or am i mistaken ? Is it related to zero copy ? What was the previous implementation for buffer sharing (i can not remember the name) ?
      Was it started by Keith Packard and is it related to DRI3 ?

      If you can point to a documentation page or wiki, i would appreciate it.

      Thanks in advance

      Comment


      • #4
        Originally posted by iznogood View Post
        Hi all,

        can someone explain why we need DMA BUF ? What problems it solves and what possible new issues it might create ? I believe it was for security reasons or am i mistaken ? Is it related to zero copy ? What was the previous implementation for buffer sharing (i can not remember the name) ?
        Was it started by Keith Packard and is it related to DRI3 ?

        If you can point to a documentation page or wiki, i would appreciate it.

        Thanks in advance
        There you go:

        Comment


        • #5
          dma-bufs are already used by mesa for egl under wayland.

          However for general needs, the protocol needs to be more complete, because a dma-buf could be created on anything, and things like the format, etc should be negociated.

          Comment


          • #6
            Originally posted by mannerov View Post
            dma-bufs are already used by mesa for egl under wayland.

            However for general needs, the protocol needs to be more complete, because a dma-buf could be created on anything, and things like the format, etc should be negociated.
            The the main point of the patches is making dma-bufs usage a lot easier and maybe the preferred solution, right?

            Comment


            • #7
              Originally posted by iznogood View Post
              Hi all,

              can someone explain why we need DMA BUF ? What problems it solves and what possible new issues it might create ? I believe it was for security reasons or am i mistaken ? Is it related to zero copy ? What was the previous implementation for buffer sharing (i can not remember the name) ?
              Was it started by Keith Packard and is it related to DRI3 ?

              If you can point to a documentation page or wiki, i would appreciate it.

              Thanks in advance
              DMA-BUFs are all about zero copy
              there wasn't such API/tool as memory block that you can use in kernel space and user space(application space) and you don't need to do memcpy of it
              every SOC company came with own implementation.
              and DMA-BUF unify this problem
              So when your hw video decoder allocate data you can use this buffer with nice api from user space in your application and then use it in you application as texture without big slowdown named memcpy
              from kernel(video decoder) -> use space(your application) -> kernel space(gpu driver) WITHOUT one single memcpy
              And you can imagine many other use cases for this

              Comment


              • #8
                Originally posted by 89c51 View Post
                Is this the first step towards being able to use multiple GPUs on WL or am i completelty confusing DMA_BUF with something else??
                This is all from memory so it might be incorrect.
                It's related, but this is about zero copy on particular. PRIME is specifically about multi device inter operability, but that requires dmabuf.

                Edit: see post above mine

                Comment

                Working...
                X