The Linaro consortium debuts

Posted Jun 4, 2010 23:10 UTC (Fri) by ai4qr (guest, #64631)
In reply to: The Linaro consortium debuts by DavidRusling
Parent article: The Linaro consortium debuts

The preferred implementation of device trees puts the data on the file system or in an accessible partition. This means that in many cases, all you'll need to enable a new device is to change a flat file, without having to reflash anything or rebuild a kernel.

The Linaro consortium debuts

Posted Jun 4, 2010 23:46 UTC (Fri) by swetland (guest, #63414) [Link]

Until you run into common scenarios like "GPIO43 must be used to enable the level shifter on the I2C bus before peripherals X, Y, and Z are enumerated" or "when inactive, you must adjust the GPIOMUX on UART3 to GPIO mode and use a gpio wakeup interrupt because you cannot wakeup from suspend on UART handshake lines without drawing too much power", etc.

I guess if your hardware is extremely rigidly specified, or you're not dealing with typical mobile/consumer device design maybe a device tree will get you going, but I'm skeptical that you'd be able to *ship* something without any modifications to the kernel (I've never seen a case of a not-completely-trivial hardware spin not requiring some annoying software change). Since I've always found that there's need for a per-product kernel build, splitting the board configuration out into some separate chunk of data has never been appealing.