19.12.2010, 15:42


Developed for and on my Flepo alpha PC, which is driven by the AMD Geode LX and its companion chip CS5536.

It's available via git in this repository under the GPLv2 license.

Up to now it can do modesetting, user-space notification of vertical blanking events and perform page-flipping. There is no support for DMA IOCTLs or the like, but the dumb buffer API for KMS is fully supported, currently through the use of the dumb memrange allocator drm_mm. Also so far it can acquire EDID data from a capable monitor connected via VGA and perform DAC load detection by driving the analog outputs with a pixel value just under half intensity and comparing the actual voltage level to a reference set to 0.35V by the manufacturer.

Although the chip supports a flat panel connector, my Flepo PC doesn't have it wired up to anything, so I haven't been able to test that circutry and so far there is no support for either flat panels or the Video Output Port (which uses the same connector), but that should change in the future, as one of the developers of the "classic" X.org-driver xf86-video-geode on #geode on FreeNode owning an OLPC XO-1, that also features an AMD Geode (and a flat panel, of course), has offered to run tests. Thanks leio.

The motivation behind this driver originated from the Pong I built for this board, featuring a (nice) variant of the 8051 microcontroller. It is driving the GPIOs to produce a valid VGA signal to be displayed on a regular monitor. Also the afore mentioned leio has pointed me to some performance issues their UMS driver currently has regarding EXA and some needless memory copy operations, that I hope to be able to alleviate through the KMS driver.

Dave Airlie has been so kind to provide a new branch called restart for xf86-video-modesetting that not only compiles (unlike the original version of the driver), but also provides support for a hardware accelerated pointer using (only!) the KMS API. Using this driver, lx can already be tested, though no acceleration other than the cursor is available through the pure KMS API. Unfortunately as the kernel's DRM API still lacks a means to communicate the available/valid cursor dimensions/depths, xf86-video-modesetting needs patching to request only a 48x64 sizes cursor buffer as the kernel driver will deny any other surface sizes (due to the lack of hardware support). However, I'm expecting this to change, as soon as an API for this has stabilized and landed in the kernel. A discussion of this matter may be found in this thread on the dri-devel mailinglist.