On Thu, 5 Jun 2008 16:39:35 +0200 Matthias Hopf <[EMAIL PROTECTED]> wrote:
> That sounds *very* interesting. I guess this was only a quick first > discussion, so there probably isn't any hard evidence (read: code or > text ;) to share... Well the idea is that we better have old userspace which is know working on new kernel space to debug kernel space then we can start tackle new userspace. Also this give you 3d from day one, well once your drm is ready. Nothings yet, i am tackling a new way to recover from lockup. > > packet to the ioctl and to take advantage of nop packet or add new unknow > > packet type transformed into nop by drm to sneak in additional information > > that will be needed latter (when using a memory manager). I don't know when > > What information would that be? The equivalence of texture / surface / > memory region numbers? > > I don't know whether abusing NOPs for that is actually the right idea, > though. Information would simply be buffer id (if there is a memory manager and nothings with current drm). So that we don't have to build a complex list of buffer relocation. I am sure adding nop is a good plan, it would be just 1 or 2 dword per packet that need relocation. As example worst case on r300 is 2 * (8(textures) + 2(zbuffer+colorbuffer) + 1(vertexbuffer)) -> 22 dwords So 22 extra dword added to the normal cmd stream, doesn't sounds like too much waste :) I have toyed with many design (you can look at radeon_ms history) about cmd buffer submission and i got the feeling that this is the easiest and cleanest way to send cmd to hw. Note that we get buffer userspace sub allocator for free with this scheme while it's a lot more complex in others scheme i tested. > > So my only advice/wishes :) is that you start from somethings like that > > ie not and ioctl where you submit packet inside a container structure > > but directly a buffer which can be parsed by hw. Maybe this can make > > cmd checking code a little bit more complexe but i don't think so. > > Why do you think it would be problematic to have a new ioctl with a > container structure? Not that I intend to do anything TTM/GEM related > ATM, but I'm curious. Just for code reuse :) if you got the properly formated way then when gem popup we will be able to pretty much cut and paste this code :) Cheers, Jerome Glisse <[EMAIL PROTECTED]> ------------------------------------------------------------------------- Check out the new SourceForge.net Marketplace. It's the best place to buy or sell services for just about anything Open Source. http://sourceforge.net/services/buy/index.php -- _______________________________________________ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
