Re: [RFC: drm server] limitations of MiG for ioctls

[Yuqian Yang]

> It sounds like we need to implement a wrapping-paper thin layer in 
> libdrm
> that unpacks drmIoctl() and fans out all the combinations of proper 
> RPCs;
> it will be ugly but doable.  (I was hoping we could just pass in a
> casted struct as that is what an ioctl is for). This is more work,
> but probably simplest.

That's what I think. Compared to rewriting the whole graphics stack 
(supposing we write something replacing drm, then we need a new mesa), 
the UGLY is more acceptable, I think.

But based on what you have done, I think it should be a little easier 
now than from scratch, as you have already migrated most structs and 
functions.