> Though the whole clk handling is quite strange. > I would assume that it's always a > - clk-request > - clk-enable > something > - clk-disable > - clk-release > > That a clk-disable also hides in the clk-release is really surprising, > and I guess a lot of places in u-boot will encounter the same > imbalance - just mitigated by the fact that the double-disable > won't matter. > > Directly in clk_get_bulk the problem is apparent ... > The function tries to get all the clocks, and on error calls > clk_release_all on the already gotten clocks. > > All of these clocks never got enabled, but are getting double-disabled > by the clk_release_all() call. > > Very very strange ;-)
Right. clk_release_bulk() hides a clk_disable(), and clk_get_bulk() does the same in its own error path: it calls clk_release_all() on clocks it only ever requested, never enabled. Everywhere else that double-disable is a no-op, so nobody notices. The gated-fixed-clock driver is what makes it bite, since there the disable actually toggles the regulator. Thanks for the review, and for testing on the rock-5-itx. Daniele
