Announcement

Collapse
No announcement yet.

NVIDIA Issues New 302 Linux Beta

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

  • #21
    Isn't KMS also required to run Xorg as non-root, due to drivers shifting into kernelspace ring (so no root priviledges for hardware-level calls are needed with KMS) ? So, KMS = tolerance against priveledge escals in Xorg.

    Also, KMS gives a bit of performance boost and frees from Xorg dependency for graphics per se.

    Nvidia linux drivers are result of cutting down windows release and refactoring it with bindings (to my knowledge), so the base driver is designed for (more-less) microkernel nt, which means it will be additional effort (for every driver release) to make it work in-kernel well, which means extra cost to nvidia, which means no.
    Last edited by crazycheese; 05-21-2012, 11:10 AM.

    Comment


    • #22
      Originally posted by crazycheese View Post
      Isn't KMS also required to run Xorg as non-root, due to drivers shifting into kernelspace ring (so no root priviledges for hardware-level calls are needed with KMS) ?
      No, nvidia hasn't needed X as root since forever, since before there was KMS. All you need is access to the /dev/nvidia* nodes. By default everyone has that, some distros (I know of Gentoo) limit access to users in the video group. The video group is also default for KMS drivers, look into /dev/dri and you'll see it. So there's no difference between open drivers and nvidia in this regard. The reason we still run X as root is input handling, not graphics.

      Originally posted by crazycheese View Post
      Also, KMS gives a bit of performance boost and frees from Xorg dependency for graphics per se.
      I recall tests showing that KMS/DRI2 actually has a bit lower performance than UMS/DRI1. I recall also reading about the reasons for that, but I don't know them right now, I'd have to search again to refresh my memory.

      Comment


      • #23
        Originally posted by Gusar View Post
        I recall tests showing that KMS/DRI2 actually has a bit lower performance than UMS/DRI1.
        Note for Michael: that one is worth new article "KMS/UMS performance comparsion".

        Comment


        • #24
          Originally posted by Gusar View Post
          And yes, you can switch between drivers without reboot. You need to unbind nouveau from the console, then you can rmmod the module. From here:
          Code:
          echo 0 > /sys/class/vtconsole/vtcon1/bind
          rmmod nouveau
          /etc/init.d/consolefont restart
          rmmod ttm
          rmmod drm_kms_helper
          rmmod drm
          Cool, thanks! Will try that.

          Comment


          • #25
            Originally posted by crazycheese View Post
            Isn't KMS also required to run Xorg as non-root, due to drivers shifting into kernelspace ring (so no root priviledges for hardware-level calls are needed with KMS) ? So, KMS = tolerance against priveledge escals in Xorg.
            kernel modesetting is required, KMS is not.

            They really should have named KMS something else, because it seems every single time KMS comes up people confuse the OSS API implementation with the general idea the acronym stands for.

            Comment


            • #26
              Originally posted by not.sure View Post
              I'm not.sure KMS was such a great idea to begin with. For example, is it possible to switch between nouveau and blob without rebooting? It seems thanks to KMS you can't ever unload the nouveau kernel module, whereas the blob is only needed to start X. That also allows for easy driver updates w/o reboot.
              Well... "easy driver updates w/o reboot" is overstating it a bit. I mean you do have to close down X, unload the blob kernel module, reload the new one, then restart X. You can do it... but rebooting is probably easier.

              Comment


              • #27
                Originally posted by smitty3268 View Post
                kernel modesetting is required, KMS is not.

                They really should have named KMS something else, because it seems every single time KMS comes up people confuse the OSS API implementation with the general idea the acronym stands for.
                Maybe it should be named KMS-API ?

                Comment


                • #28
                  Originally posted by johnc View Post
                  Well... "easy driver updates w/o reboot" is overstating it a bit. I mean you do have to close down X, unload the blob kernel module, reload the new one, then restart X. You can do it... but rebooting is probably easier.
                  I don't think that 'not.sure' was overstating it at all. Going from init 5 to init3, unload/reload nvidia, then switch back to init 5 is not a big deal, nor difficult to do and is many times faster than rebooting.

                  I would argue that rebooting is only easier if the user doesn't really know what i would call "linux basics" and/or they are uncomfortable with maintaining their own system.

                  Comment


                  • #29
                    Originally posted by ninez View Post
                    I don't think that 'not.sure' was overstating it at all. Going from init 5 to init3, unload/reload nvidia, then switch back to init 5 is not a big deal, nor difficult to do and is many times faster than rebooting.

                    I would argue that rebooting is only easier if the user doesn't really know what i would call "linux basics" and/or they are uncomfortable with maintaining their own system.
                    How can I do this then? Because I get nvidia driver updates frequently and if there's an easier way, I'd want to do it.

                    The way I was doing it was to log out of my gnome session, go to a VT, login, kill all gnome and X stuff, rmmod the module, insmod the new one, then restart X and gnome, logout on the VT console then go back to my X VT and login. Is there an easier way?

                    In comparison, rebooting takes about 15 seconds for me and requires no input after selecting the option. The biggest neg to the reboot is that I lose my open windows and have to login and restart everything, but that seems necessary regardless.

                    Maybe on the same topic... I'm kinda surprised that kernel hot-swapping isn't available yet. Haven't they been talking about this for years?

                    Comment


                    • #30
                      Originally posted by johnc View Post
                      How can I do this then? Because I get nvidia driver updates frequently and if there's an easier way, I'd want to do it.

                      The way I was doing it was to log out of my gnome session, go to a VT, login, kill all gnome and X stuff, rmmod the module, insmod the new one, then restart X and gnome, logout on the VT console then go back to my X VT and login. Is there an easier way?
                      why logout,kill gnome, xorg, and switch VT (all) manually? - that doesn't make any sense to me. You know what runlevels are right?

                      http://en.wikipedia.org/wiki/Runlevel

                      Instead of wasting your time ~ open gnome-terminal and type: 'sudo init 3' - which will do all of that in one fowl swoop, minus switching VTs which isn't even needed in this case.... then once you have unloaded / reloaded (new) nvidia... then just type; sudo init 5 (which will start up everything again) ~ that's what i mean about linux basics, understanding what something as old and common place as runlevels. what they are and how to use them, properly.

                      you are doing a bunch of unnecessary stuff, that probably does make rebooting seem easier.

                      Originally posted by johnc View Post
                      In comparison, rebooting takes about 15 seconds for me and requires no input after selecting the option. The biggest neg to the reboot is that I lose my open windows and have to login and restart everything, but that seems necessary regardless.
                      yup, necessary regardless

                      Originally posted by johnc View Post
                      Maybe on the same topic... I'm kinda surprised that kernel hot-swapping isn't available yet. Haven't they been talking about this for years?
                      I think so, but honestly don't pay that much attention.
                      Last edited by ninez; 05-21-2012, 05:02 PM.

                      Comment

                      Working...
                      X