On Fri, 14 Apr 2023 at 20:48, Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote: > > > > On 14.04.2023 18:31, Dmitry Baryshkov wrote: > > On Fri, 14 Apr 2023 at 14:26, Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote: > >> > >> Add support for the Video Clock Controller found on the SM8350 SoC. > >> > >> Signed-off-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> > >> --- > > [...] > > >> +static struct clk_rcg2 video_cc_ahb_clk_src = { > >> + .cmd_rcgr = 0xbd4, > >> + .mnd_width = 0, > >> + .hid_width = 5, > >> + .parent_map = video_cc_parent_map_0, > >> + .freq_tbl = ftbl_video_cc_ahb_clk_src, > >> + .clkr.hw.init = &(const struct clk_init_data) { > >> + .name = "video_cc_ahb_clk_src", > >> + .parent_data = video_cc_parent_data_0, > >> + .num_parents = ARRAY_SIZE(video_cc_parent_data_0), > >> + .flags = CLK_SET_RATE_PARENT, > >> + .ops = &clk_rcg2_shared_ops, > >> + }, > >> +}; > > > > Do we need this clock at all? We don't have the child > > video_cc_ahb_clk, so potentially CCF can try disabling or modifying > > this clock. > Hm.. I see a few things: > > 1. downstream kona has it, upstream does not > 2. it's shared so we never actually hard-shut it off.. > 2a. ..but it'd be good to ensure it's on when it's ready.. > 2b. ..but we never do anyway.. > 2c. ..but should we even? doesn't Venus govern it internally? > > > > > >> + > >> +static const struct freq_tbl ftbl_video_cc_mvs0_clk_src[] = { > >> + F(720000000, P_VIDEO_PLL0_OUT_MAIN, 1, 0, 0), > >> + F(1014000000, P_VIDEO_PLL0_OUT_MAIN, 1, 0, 0), > >> + F(1098000000, P_VIDEO_PLL0_OUT_MAIN, 1, 0, 0), > >> + F(1332000000, P_VIDEO_PLL0_OUT_MAIN, 1, 0, 0), > >> + { } > >> +}; > >> + > > [...] > > >> +static struct clk_branch video_cc_mvs1_clk = { > >> + .halt_reg = 0xdb4, > >> + .halt_check = BRANCH_HALT_VOTED, > > > > As a note, sm8250 has BRANCH_HALT here. > No, it does on the div2 clk, and so do we: Excuse me, I got confused by all the syllables. I was looking at the video_cc_mvs1c_clk. On sm8250 it is _VOTED, in this patch it is not. I can not say that either one of those is incorrect, but such a difference looks a bit suspicious for me. Maybe Tanya or somebody else can comment here. > [...] > > >> +}; > >> + > >> +static struct clk_branch video_cc_mvs1_div2_clk = { > >> + .halt_reg = 0xdf4, > >> + .halt_check = BRANCH_HALT_VOTED, > >> + .hwcg_reg = 0xdf4, > > [...] > > >> + > >> +static const struct qcom_reset_map video_cc_sm8350_resets[] = { > >> + [CVP_VIDEO_CC_INTERFACE_BCR] = { 0xe54 }, > >> + [CVP_VIDEO_CC_MVS0_BCR] = { 0xd14 }, > > > > Would it be better to use common VIDEO_CC prefix here (IOW: > > VIDEO_CC_CVP_MVS0_BCR, VIDEO_CC_CVP_INTERFACE_BCR), etc. > My best guess would be that the ones prefixed with CVP_ > are actual INTF/INSTANCEn(CORE) reset lines whereas > the ones containing _CLK_ reset their clock sub-branches. Note, again, on sm8250 all resets start with VIDEO_CC, even CVP ones. I think we can follow that. > > > > >> + [VIDEO_CC_MVS0C_CLK_ARES] = { 0xc34, 2 }, > >> + [CVP_VIDEO_CC_MVS0C_BCR] = { 0xbf4 }, > >> + [CVP_VIDEO_CC_MVS1_BCR] = { 0xd94 }, > >> + [VIDEO_CC_MVS1C_CLK_ARES] = { 0xcd4, 2 }, > >> + [CVP_VIDEO_CC_MVS1C_BCR] = { 0xc94 }, > >> +}; > > [...] > > >> + ret = pm_runtime_resume_and_get(&pdev->dev); > >> + if (ret) > >> + return ret; > >> + > >> + regmap = qcom_cc_map(pdev, &video_cc_sm8350_desc); > >> + if (IS_ERR(regmap)) { > >> + pm_runtime_put(&pdev->dev); > >> + return PTR_ERR(regmap); > >> + }; > > > > Extra semicolon > Ooeh! > > > > >> + > >> + clk_lucid_pll_configure(&video_pll0, regmap, &video_pll0_config); > >> + clk_lucid_pll_configure(&video_pll1, regmap, &video_pll1_config); > >> + > >> + /* > >> + * Keep clocks always enabled: > >> + * video_cc_ahb_clk > >> + * video_cc_xo_clk > >> + */ > >> + regmap_update_bits(regmap, 0xe58, BIT(0), BIT(0)); > >> + regmap_update_bits(regmap, 0xeec, BIT(0), BIT(0)); > >> + > >> + ret = qcom_cc_really_probe(pdev, &video_cc_sm8350_desc, regmap); > >> + pm_runtime_put(&pdev->dev); > >> + > >> + return ret; > >> +} > >> + > >> +static const struct dev_pm_ops video_cc_sm8350_pm_ops = { > >> + SET_RUNTIME_PM_OPS(pm_clk_suspend, pm_clk_resume, NULL) > > > > The driver doesn't use pm_clk at all. Are these PM_OPS correct? > I'm unsure. I see the pm state changing in debugfs when the clocks are > (not) consumed. But let's continue our discussion about using pm_clks > for AHB. Well, those are two separate questions. One is that w/o additional pm_clk calls this string is useless (and should be removed). Another on is a possible restructure of our cc drivers to use pm_clk for AHB clocks (which would require adding more than that). > > > > >> +}; > >> + > >> +static const struct of_device_id video_cc_sm8350_match_table[] = { > >> + { .compatible = "qcom,sm8350-videocc" }, > >> + { } > >> +}; > >> +MODULE_DEVICE_TABLE(of, video_cc_sm8350_match_table); > >> + > >> +static struct platform_driver video_cc_sm8350_driver = { > >> + .probe = video_cc_sm8350_probe, > >> + .driver = { > >> + .name = "sm8350-videocc", > >> + .of_match_table = video_cc_sm8350_match_table, > >> + .pm = &video_cc_sm8350_pm_ops, > >> + }, > >> +}; > >> +module_platform_driver(video_cc_sm8350_driver); > >> + > >> +MODULE_DESCRIPTION("QTI SM8350 VIDEOCC Driver"); > >> +MODULE_LICENSE("GPL"); > >> > >> -- > >> 2.40.0 > >> > > > > Generic note: the register layout follows closely sm8250. However the > > existing differences probably do not warrant merging them. > No, I don't think merging any designs that are farther away > than 8150 and 8155 or 8992 and 8994 etc. is a good idea.. > > I don't want to ever look at something like dispcc-sm8[123]50.c > again! Me too! -- With best wishes Dmitry