Commit 2702ea7dbec5 ("i2c: designware: wait for disable/enable only if necessary") seems to break things on my (Bay Trail) Teclast x89 HD tablet. Since kernel 4.9, and still in 4.11-rc3, my (goodix) touchscreen connected over i2c misses presses and/or release events which makes it basically unusable. Reverting commit 2702ea7dbec5 in the 4.9 kernel does fix the issue. However, reverting this commit in the 4.10 kernel does *not* fix the issue... so either this is a heisenbug, or there are two different bugs instead of one. I'm trying to bisect the issue, between 4.9 (with 2702ea7dbec5 reverted) and 4.10, but most of the kernels I build between these two versions fail to boot, which makes things very difficult...
Could you try how does v4.9.17 work with the revert? One difference between v4.9 and v4.9.17 is that commit 12688dc21f71 ("Revert "i2c: designware: detect when dynamic tar update is possible"") got reverted during v4.10 and was sent back to v4.9 stable kernels too.
Kernel v4.9.17 works fine with the revert, just like 4.9.0, so the commit you mention doesn't seem to be involved.
Both patches were tested in a Bay Trail platform(Minnowboard Max) and Cherry Trail(Intel Aero) and worked with great performance improvements, did you have debug what is failing? There is any error returned after a I2C transaction?
Bug found and fix sent upstream: https://patchwork.kernel.org/patch/9654233/
Nice :D
Should have landed by v4.14-rc6. Can be closed then?