3 ms·

Linus' response: We'll see. I'm cautiously optimistic that this is a real shift in how Nvidia perceives Linux. The actual docs released so far are fairl

by kunai 13y ago

Linus' response: We'll see. I'm cautiously optimistic that this is a real shift in how Nvidia perceives Linux. The actual docs released so far are fairly limited, and in themselves they wouldn't be a big thing, but if Nvidia really does follow up and start opening up more, that would certainly be great.

cobrausn 13y ago
Why not include the rest? Or at least, the rest of what was mentioned in the article. They've already been much better in the ARM SoC space than they were on the more traditional GPU side, and I really hope that some day I can just apologize for ever giving them the finger.
- sz4kerto 13y ago
  BTW, that difference comes from the different amount of proprietary IP in the respective fields.
  - ajross 13y ago
    There really is comparatively little "proprietary IP" on the kernel side of a GPU driver. In almost all cases, the kernel is responsible for doing things like setting modes, managing memory allocation/mapping between the GPU and CPU domains, and transmitting command buffers from userspace to the hardware. Where there is proprietary stuff, it's generally related to stuff like DRM (HDCP) or video playback which aren't covered by the kernel DRM framework anyway and wouldn't prevent an free interoperable driver.
    - Arelius 13y ago
      I mean, glsl compilers and optimizers. And as time goes on, Nvidia is moving more of the fixed function out of silicon into GPU software. Which is effectively binary blobs that need to get uploaded to the GPU. This all seems like quite a bit of "propietary IP"
    - rys 13y ago
      There is a lot of proprietary IP in a modern GPU driver: https://news.ycombinator.com/item?id=6176498 https://news.ycombinator.com/item?id=6176498
      - ajross 13y ago
        Shrug; you're talking about reverse engineering hardware implementation from the details of the transferred data. Source access is hardly required for that, and the Nouveau driver we're talking about is an existence proof. And in any case all modern GPU manufacturers except NVIDIA ship GPL kernel modules and aren't leaking "proprietary IP" to each other. This isn't an issue of secret sauce in the kernel, sorry. NVIDIA just don't want to spend the effort to conform to someone else's architecture.
        swetland 13y ago
        Even NVIDIA takes this approach on their mobile integrated GPUs on Tegra (GPL'd kernel driver). I wish they'd do so on the desktop too. Having the kernel driver (which really does not need to be all that much code) be open source makes it massively easier to upgrade the kernel around the GPU driver.

by kunai 13y ago

cobrausn 13y ago

Why not include the rest? Or at least, the rest of what was mentioned in the article. They've already been much better in the ARM SoC space than they were on the more traditional GPU side, and I really hope that some day I can just apologize for ever giving them the finger.

sz4kerto 13y ago
BTW, that difference comes from the different amount of proprietary IP in the respective fields.
- ajross 13y ago
  There really is comparatively little "proprietary IP" on the kernel side of a GPU driver. In almost all cases, the kernel is responsible for doing things like setting modes, managing memory allocation/mapping between the GPU and CPU domains, and transmitting command buffers from userspace to the hardware. Where there is proprietary stuff, it's generally related to stuff like DRM (HDCP) or video playback which aren't covered by the kernel DRM framework anyway and wouldn't prevent an free interoperable driver.
  - Arelius 13y ago
    I mean, glsl compilers and optimizers. And as time goes on, Nvidia is moving more of the fixed function out of silicon into GPU software. Which is effectively binary blobs that need to get uploaded to the GPU. This all seems like quite a bit of "propietary IP"
  - rys 13y ago
    There is a lot of proprietary IP in a modern GPU driver: https://news.ycombinator.com/item?id=6176498 https://news.ycombinator.com/item?id=6176498
    - ajross 13y ago
      Shrug; you're talking about reverse engineering hardware implementation from the details of the transferred data. Source access is hardly required for that, and the Nouveau driver we're talking about is an existence proof. And in any case all modern GPU manufacturers except NVIDIA ship GPL kernel modules and aren't leaking "proprietary IP" to each other. This isn't an issue of secret sauce in the kernel, sorry. NVIDIA just don't want to spend the effort to conform to someone else's architecture.
      - swetland 13y ago
        Even NVIDIA takes this approach on their mobile integrated GPUs on Tegra (GPL'd kernel driver). I wish they'd do so on the desktop too. Having the kernel driver (which really does not need to be all that much code) be open source makes it massively easier to upgrade the kernel around the GPU driver.