On 2016/8/26 23:46, Doug Anderson wrote: > Hi, > > On Fri, Aug 26, 2016 at 1:31 AM, Heiko St?bner <heiko at sntech.de> wrote: >> The aclk_emmc_grf is (similar to aclk_vio_grf etc) part of the emmc block. > > Right now aclk_emmc_grf is set as CLK_IGNORE_UNUSED, which is my least > favorite clock property. ;) > > Due to the parenting structure and the fact that we just added > "aclk_emmc_noc" as a critical clock, I believe that this means that > aclk_emmc_grf will never turn off. It would be awfully nice if you It just means we don't gate aclk_emmc_grf/aclk_emmc_noc if not referenced, but actually we don't prevent clk driver to gate its parent which means it isn't really CLK_IGNORE_UNUSED or critical to a certain degree from my personal view as we may need to propagate this feature forward to its parents. Anyway, that seems a bit beyond the topic of what $SUBJECT wanna handle with. But I agree that we should make this dependency relationship more exlicit to users anyway. I will do it once $SUBJECT applied. :) Thanks. > could send up a patch to actually turn the "grf" clock on when needed > and turn it off when not needed. That would let us remove the > "CLK_IGNORE_UNUSED" from all of the eMMC clocks in rk3399. > > I noticed you send v2 of your patch series and that looks fine to me. > The issue with not handling the "grf" clock is not a new one and > shouldn't block your patch. ...but it would still be nice to handle > properly. > > -Doug > > > -- Best Regards Shawn Lin
