On 26/11/12 16:58, Tarl Neustaedter wrote:
Now AFAICT one of two things is wrong here:
- get-package-path should be fixed so that if a phandle belongs to a
leaf node, the unit address is not included as part of the node name
- open-dev should happily accept the trailing @8 on the keyboard node
Does anyone know which one of these is the correct solution?
What does "adb" have for encode-unit and decode-unit? The fact that you can open "keyboard" but not "keyboard@8" indicates that the encode-unit/decode-unit aren't properly translating between reg property and string unit-address.
I've finally worked out what was going on here. The adb device doesn't have explicit encode-unit/decode-unit words, and currently if the path-resolution algorithm in OpenBIOS can't find one then it switches over to a default version based upon a single unit cell.
The reason that opening the keyboard with the trailing @8 didn't work was because there was no symmetry; i.e. if decode-unit didn't exist, the process terminated without matching rather than translating the single unit cell back into a string.
I now have a working patch to implement the opposite for decode-unit which doesn't cause regressions on any of my test images, and allows direct opening of the keyboard device obtained from get-package-path without any extra string manipulation. So unless I find any other issues, I'll apply this along with a correction for the previous patch over the next couple of days.