On Sunday 11 January 2009, Stijn Devriendt wrote: > TI told us not to use some of their drivers as modules because they break. > We haven't investigated since we'll use these drivers built-in to the > kernel anyway, > but looking at the initialization code, I had fair reasons to suspect > the pinmux layer.
So TI told you they weren't fixing known bugs in some of their drivers? I suppose they have other priorities; but bugs tend to hide other bugs... > > Near as I can tell, the dm6467 stuff is barely working > > in the DaVinci GIT tree. The goal is to have G.O.O.D. > > code there, ready to go to mainline. If it's broken, > > it needs fixing.... > > Well, you're more or less convincing me to step forward > and improve the code quality. In some way I'm already doing > that. Good. As you've noticed, there are bumps in the process of making robust drivers be properly available. For the dm6467 in particular, chip support wasn't even available in that GIT tree until quite recently. > As an MV customer I've sent in several bug reports that > they are getting fixed. Assuming MV shares code of course ;) > > We're integrating our custom design platform with the DM6467 at > its heart and it was a struggle with bad code and a minimalistic > approach from TI's side. I can't say more about TI's motivations than just observe that they're a hardware company, not a software company. Also that maybe it's not entirely in the interest of folk like MontaVista for you not to need their help. ;) > Why does adding a serial flash and using another > ethernet PHY than what's on the EVM need to be such a pain. There > are clean interfaces in the linux kernel to make it all work seamlessly > and then you see that it's just not being used. Some of that trouble is inherent in using older kernels, and non-mainline code (both platform support and drivers) that hasn't been brought up to community standards. > (although I saw a patch > passing last or this month which adds PHY-layer support). Yeah, I suspect the davinci_emac.c driver may nearly be ready to submit for mainline. Switch over to ioremap() and most of the obvious structural goofage is gone ... a review cycle on netdev will turn up other things to update before merging. > Apart from that I've been looking to fetch the git-tree from > source.mvista.com/git > this weekend, but found a dead URI instead. Yeah, sometimes that GIT server seems to flake out. I noticed the same thing when trying to refresh a workspace the other day. - Dave _______________________________________________ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
