Re: [Dri-devel] video-reset and DRM

[Alan C. <al...@lx...>]

On Mer, 2004-03-17 at 19:40, Jon Smirl wrote:
> secondary cards if needed. The patch works by generating a hotplug event on
> driver insertion. This event calls a user space helper - video-reset. Video
> cards can be reset from user space using VM86 mode.

Sometimes. Many of these drivers require substantial emulation
frameworks or use stuff like DMA and IRQ lines. You also have to
intercept and virtualise PCI access in the vm86 environment to get the
PCI bus locking right. Some of it is doable, other stuff like 3dfx
nobody ever got to work


> The VBIOS ROM is retrieved using standard kernel calls for enabling/disabling
> it. Right now X enables it from user space without telling the kernel. This
> leaves a window of vulnerability where the kernel could assign another PCI
> device on top of the video ROM (hotplug) since the kernel doesn't know it has
> been enabled.

The Xserver maps the ROM over the address space used by one of the
existing BARs. They thought of that 8)

> To me the video-reset program is an extension of the DRM device driver and
> should be kept in the new drm tree. But other people have different opinions.
> What does everyone think?

Certain ACPI things need video setup too, and btw I for one agree with
you about something - for some cards the video reset being handled via
hotplug actually may make sense.