On Tue, Mar 1, 2016 at 11:59 PM, Wolfram Sang [off-list ref] wrote:

quoted

+     ret = clk_prepare_enable(i2c->clk);
+     if (ret)
+             dev_err(&pdev->dev, "Unable to enable clock.\n");

Don't you want to bail out here? The clk refcounting will be messed up
if you continue.

Agree missed it will fix it.

quoted

+err_clk_dis:
+     clk_disable_unprepare(i2c->clk);
+     pm_runtime_set_suspended(&pdev->dev);
+     pm_runtime_disable(&pdev->dev);
+     return ret;

To match the inverse of the activation order, I'd put the
clk_disable_unprepare after the pm_* calls. Couldn't it happen that
those calls need to change the clk state?

makes sense will change in the next version