On 2/28/23 12:19, Javier Martinez Canillas wrote: > Gerd Hoffmann <kraxel@xxxxxxxxxx> writes: > > Hello Gerd, > >> On Mon, Feb 27, 2023 at 07:40:11AM -0800, Rob Clark wrote: >>> On Sun, Feb 26, 2023 at 10:38 PM Gerd Hoffmann <kraxel@xxxxxxxxxx> wrote: >>>> On Fri, Feb 24, 2023 at 10:02:24AM -0800, Rob Clark wrote: >>>>> From: Rob Clark <robdclark@xxxxxxxxxxxx> >>>>> >>>>> Add a build option to disable modesetting support. This is useful in >>>>> cases where the guest only needs to use the GPU in a headless mode, or >>>>> (such as in the CrOS usage) window surfaces are proxied to a host >>>>> compositor. >>>> Why make that a compile time option? There is a config option for the >>>> number of scanouts (aka virtual displays) a device has. Just set that >>>> to zero (and fix the driver to not consider that configuration an >>>> error). >>> The goal is to not advertise DRIVER_MODESET (and DRIVER_ATOMIC).. I >>> guess that could be done based on whether there are any scanouts, but >>> it would mean making the drm_driver struct non-const. >> Apparently there is a drm_device->driver_features override, >> (amdgpu uses that). The driver could simply drop the DRIVER_MODESET and >> DRIVER_ATOMIC bits in case no scanout is present instead of throwing an >> error. >> >>> And I think it is legitimate to allow the guest to make this choice, >>> regardless of what the host decides to expose, since it is about the >>> ioctl surface area that the guest kernel exposes to guest userspace. >> I think it is a bad idea to make that a compile time option, I'd suggest >> a runtime switch instead, for example a module parameter to ask the >> driver to ignore any scanouts. >> > I don't think there's a need for a new module parameter, there's already > the virtio-gpu 'modeset' module parameter to enable/disable modsetting > and the global 'nomodeset' kernel cmdline parameter to do it for all DRM > drivers. > > Currently, many drivers just fail to probe when 'nomodeset' is present, > but others only disable modsetting but keep the rendering part. In fact, > most DRM only drivers just ignore the 'nomodeset' parameter. IIUC, Rob's main point for having a config option is solely for security reasons. The config option eliminates possibility of accidentally (or intentionally) enabling KMS from software, which is better to have in case of shipping a product (Chromebook) on which multiple teams are working on. -- Best regards, Dmitry