| Devicetree (DT) ABI | |
| I. Regarding stable bindings/ABI, we quote from the 2013 ARM mini-summit | |
| summary document: | |
| "That still leaves the question of, what does a stable binding look | |
| like? Certainly a stable binding means that a newer kernel will not | |
| break on an older device tree, but that doesn't mean the binding is | |
| frozen for all time. Grant said there are ways to change bindings that | |
| don't result in breakage. For instance, if a new property is added, | |
| then default to the previous behaviour if it is missing. If a binding | |
| truly needs an incompatible change, then change the compatible string | |
| at the same time. The driver can bind against both the old and the | |
| new. These guidelines aren't new, but they desperately need to be | |
| documented." | |
| II. General binding rules | |
| 1) Maintainers, don't let perfect be the enemy of good. Don't hold up a | |
| binding because it isn't perfect. | |
| 2) Use specific compatible strings so that if we need to add a feature (DMA) | |
| in the future, we can create a new compatible string. See I. | |
| 3) Bindings can be augmented, but the driver shouldn't break when given | |
| the old binding. ie. add additional properties, but don't change the | |
| meaning of an existing property. For drivers, default to the original | |
| behaviour when a newly added property is missing. | |
| 4) Don't submit bindings for staging or unstable. That will be decided by | |
| the devicetree maintainers *after* discussion on the mailinglist. | |
| III. Notes | |
| 1) This document is intended as a general familiarization with the process as | |
| decided at the 2013 Kernel Summit. When in doubt, the current word of the | |
| devicetree maintainers overrules this document. In that situation, a patch | |
| updating this document would be appreciated. |