On Thursday 01 June 2006 01:34, Simon Baxter wrote: > > I am sure following since I have nearly the same problem on my SuSE 10.1 > > system (see my posting "ivtv-0.6.x / gcc-4.1 / SUSE 10.1 problems" of > > 2006-05-24) for which there also have been no helpful replies. > > The only thing I know is that the problem is depended on the > > compiler/linker toolchain. I can only think of two reasons at > > the moment: > > > > 1) a code generation bug in gcc-4.1. > > 2) better code optimization in gcc-4.1 breaks some assumptions of > > the i2c code. The compiler has really no idea about the dependencies > > needed for accessing hardware (e.g a specific store that has to > > occur before a load) and may reorder C level statements at will as > > long as the C semantics are preserved. > > > > Given that everything else compiled by the gcc 4.1 looks to be fine > > and how often I wrongly tried to blame a compiler for some bugs I > > tend to favor 2). > > How did you arrive at this conclusion? Well after a lot of tries to get ivtv to play nice on my new 10.1 installation including various permutions of 2.6.16.x kernels and ivtv versions I finally tried to compile a known working configuration (the one I used with SUSE 10.0) on 10.1 (by restoring the src from my 10.0 backup) and compiled it with identical configuration with the 10.1 toolchain. The results were identical to my other tries compiled under 10.1 (not working).
Since the kernel (and its modules) have nearly no external dependencies except the compiler toolchain there is not much left I can think off which could be the cause. The one thing which surprises me at the moment is that according to your description your ivtv installation seems to work at least sometimes. For me it is either working (gcc-4.0 compile) or not (gcc-4.1 compile). Thomas _______________________________________________ ivtv-devel mailing list [email protected] http://ivtvdriver.org/mailman/listinfo/ivtv-devel
