[Intel-gfx] _PIXEL_CLOCK define in edid.h (GM45 chipset)

Alan DuBoff aland at softorchestra.com
Fri Feb 26 20:24:25 CET 2010


I have an odd problem in trying to build an xorg intel_drv.so 
driver to support the GM45 which is on an embedded device.

I'm trying to get MontaVisa Linux 6 working on a device, which 
is not really an embedded device per se, it has 4gig of memory 
and a solid state hard drive, but the company has asked me to 
get MVL 6 running on it. This requires that I cross compile with 
the Monta Vista tools in order to support their libraries.

MVL 6 ships with version 2.6.28r10 of the Linux kernel. On 
the distribution, Xorg is at 1.5.3, and the Intel module version 
= 2.4.2.

If I load the glx and dri driver on this, I get the exact same 
symptoms that Keith Packard describes at the following link, a 
black screen that appears to hang the system. It is not 
completly hung, I can ssh into the device and get the Xorg log.

https://kerneltrap.org/mailarchive/linux-kernel/2008/10/20/3727534/thread

This is the exactly erro I get in the Xorg.0.log with both glx 
and dri loaded with the 2.4.2 module.

--------------

(II) intel(0): [DRI] installation complete
(II) intel(0): xf86BindGARTMemory: bind key 0 at 0x01f7f000 
(pgoffset 8063)
(WW) intel(0): xf86BindGARTMemory: binding of gart memory with 
key 0
         at offset 0x1f7f000 failed (Invalid argument)

Fatal server error:
Couldn't bind memory for HW status


Backtrace:
0: /usr/bin/X(xorg_backtrace+0x38) [0x8131428]

FatalError re-entered, aborting
Caught signal 11.  Server aborting

--------------

Through some searching, I found on the intelgraphics site, there 
is a version 2.6.0 which looks to have several fixes, and it 
seems it will run on the Linux kernel 2.6.28r6+. This is from 
the Q42008 drop.

I was able to cross compile libdrm at 2.4.4, but when I try to 
get the xf86-video-intel-2.6.0 package compiled, I seem to be 
missing _PIXEL_CLOCK in edid.h. In fact, the versions of edid.h 
I have are very sparce, both on MVL6, OpenSuSE 11.1, and Debian 
Lenny (stable).

Yet, I find this version of edid.h when googling on the web, 
which appears to be Apple's version of it, and it does in fact 
have _PIXEL_CLOCK defined.

http://www.opensource.apple.com/source/X11server/X11server-85/Xquartz/xorg-server-1.4.2-apple45/hw/xfree86/ddc/edid.h

Is Intel expecting to pick up that define from edid.h?

Or what is the proper way to get the driver compiled so that I 
can run it on MVL6?

As an additional piece of data, I can load the glx without dri 
(no accelleration) and not hang the X server, but that doesn't 
help us much as we would end up with a non-accellerated driver. 
The problem seems to be memory related to the dri driver.

--

Alan DuBoff - Software Orchestration
http://www.softorchestra.com:8080/roller/blog/







More information about the Intel-gfx mailing list