On 4/27/18 2:35 AM, Nathan Chancellor wrote:
That's not good. Have you reported this to the gpio developers and authors of that patch?
I have not although I've gone ahead and CC'd them on this email.
I'm happy to carry the revert in my tree given it is probably a correct change in mainline but I figured I would let you guys know so you can make that decision.
Ah, do you think this is a userspace issue with your device and not a kernel problem? That implies that this patch isn't ok and should be reverted upstream.
I don't think this is a userspace regression. My guess is certain out of tree drivers expect a static base value, based on the wording of the commit message. Unfortunately, his is all a little bit over my head at the moment. I will try to study up on it tomorrow. I just wanted to report the regression as soon as I uncovered it (I'd be lying if I said I liked reporting bugs heh).
Apparently, user-space should no longer be depending on any base value. Instead, gpiolib should be used. I'm guessing that the Pixel 2 XL does not use gpiolib, since gpiolib didn't exist until Linux 4.8.
So I think that this patch should be reverted in 4.4.
When the patch was submitted, I proposed an alternative that would preserve existing platforms but still support gpiolib and newer systems with multiple TLMMs:
https://lkml.org/lkml/2018/2/7/360
I thought it was a good compromise, but it was rejected.