Hi Adam, >> + /* Enable upstream clocks */ >> + ret = clk_enable(&priv->clk_apb); >> + if (ret) >> + goto dis_bus_pd; >> + >> + ret = clk_enable(&priv->clk_axi); >> + if (ret) >> + goto dis_apb_clk; >> + >> + /* Enable blk-ctrl clock to allow reset to propagate */ >> + ret = clk_enable(clk); > > I am only offering a suggestion if you ever need to push another rev. > It appears that all the clocks are either on or off and if any fail, > you immediately turn off any clocks that were enabled. There are some > bulk clock options where you can basically tell the driver to turn > them all on or all off. > > clk_get_bulk, clk_enable_bulk, clk_disable_bulk, and clk_release_all > are the functions I am thinking of. Your imx8mp_mediamix_priv > structure could then replace all the individual clocks with struct > clk_bulk. > > That might shrink the overall code and simplify the readability.
These bulk functions only work if you get all clocks from a list. We currently do not want all the clocks (simply because some are useless in this case and also because they have not yet been defined). The clks property define all these: "apb", "axi", "cam1", "cam2", "disp1", "disp2", "isp", "phy"; >> + if (ret) >> + goto dis_axi_clk; >> + setbits_le32(priv->base + BLK_CLK_EN, reset); >> + >> + /* Power up upstream GPC domain */ >> + ret = power_domain_on(domain); >> + if (ret) >> + goto dis_lcdif_clk; >> + >> + /* Wait for reset to propagate */ >> + udelay(5); > > On Linux, the GPCv2 driver was updated to increase a delay from 5 to > 10 uSeconds [1] . I don't know if that's needed here, but it's > something to consider. Thanks a lot for the pointer, I will integrate this change, even if it is experimental, let's spare ourselves a round of ghost bug hunting. > adam > > [1] - > https://web.git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/drivers/pmdomain/imx?h=next-20250403&id=2379fb937de5333991c567eefd7d11b98977d059 Thanks! Miquèl
