-----Original Message----- From: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx> Sent: Saturday, February 12, 2022 1:23 AM To: bjorn.andersson@xxxxxxxxxx; Prasad Malisetty (Temp) (QUIC) <quic_pmaliset@xxxxxxxxxxx> Cc: Andy Gross <agross@xxxxxxxxxx>; Stanimir Varbanov <svarbanov@xxxxxxxxxx>; Lorenzo Pieralisi <lorenzo.pieralisi@xxxxxxx>; Rob Herring <robh+dt@xxxxxxxxxx>; Krzysztof Wilczy??ski <kw@xxxxxxxxx>; Michael Turquette <mturquette@xxxxxxxxxxxx>; Stephen Boyd <swboyd@xxxxxxxxxxxx>; Bjorn Helgaas <bhelgaas@xxxxxxxxxx>; Prasad Malisetty <pmaliset@xxxxxxxxxxxxxx>; Vinod Koul <vkoul@xxxxxxxxxx>; linux-arm-msm@xxxxxxxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx; linux-clk@xxxxxxxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx Subject: Re: [PATCH v2 03/11] clk: qcom: gdsc: add support for clocks tied to the GDSC On 05/02/2022 01:05, Bjorn Andersson wrote: > On Fri 04 Feb 08:46 CST 2022, Dmitry Baryshkov wrote: > >> On newer Qualcomm platforms GCC_PCIE_n_PIPE_CLK_SRC should be >> controlled together with the PCIE_n_GDSC. The clock should be fed >> from the TCXO before switching the GDSC off and can be fed from >> PCIE_n_PIPE_CLK once the GDSC is on. >> >> Since commit aa9c0df98c29 ("PCI: qcom: Switch pcie_1_pipe_clk_src >> after PHY init in SC7280") PCIe controller driver tries to manage >> this on it's own, resulting in the non-optimal code. Furthermore, if >> the any of the drivers will have the same requirements, the code >> would have to be dupliacted there. >> >> Move handling of such clocks to the GDSC code, providing special GDSC >> type. >> > > As discussed on IRC, I'm inclined not to take this, because looks to > me to be the same situation that we have with all GDSCs in SM8350 and > onwards - that some clocks must be parked on a safe parent before the > associated GDSC can be toggled. > > Prasad, please advice on what the actual requirements are wrt the > gcc_pipe_clk_src. When does it need to provide a valid signal and when > does it need to be parked? [Excuse me for the duplicate, Prasad's email was bouncing] Prasad, any comments? > > Regards, > Bjorn > Hi Dmitry, Greetings !!! Sorry for the inconvenience, there was an issue with my mail so I couldn’t receive the updates properly. Now issue is resolved. I am in discussion with internal team to know more about this. I will update my comments after this discussion. Thanks -Prasad >> Cc: Prasad Malisetty <pmaliset@xxxxxxxxxxxxxx> >> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx> >> --- >> drivers/clk/qcom/gdsc.c | 41 +++++++++++++++++++++++++++++++++++++++++ >> drivers/clk/qcom/gdsc.h | 14 ++++++++++++++ >> 2 files changed, 55 insertions(+) >> >> diff --git a/drivers/clk/qcom/gdsc.c b/drivers/clk/qcom/gdsc.c index >> 7e1dd8ccfa38..9913d1b70947 100644 >> --- a/drivers/clk/qcom/gdsc.c >> +++ b/drivers/clk/qcom/gdsc.c >> @@ -45,6 +45,7 @@ >> #define TIMEOUT_US 500 >> >> #define domain_to_gdsc(domain) container_of(domain, struct gdsc, >> pd) >> +#define domain_to_pipe_clk_gdsc(domain) container_of(domain, struct >> +pipe_clk_gdsc, base.pd) >> >> enum gdsc_status { >> GDSC_OFF, >> @@ -549,3 +550,43 @@ int gdsc_gx_do_nothing_enable(struct generic_pm_domain *domain) >> return 0; >> } >> EXPORT_SYMBOL_GPL(gdsc_gx_do_nothing_enable); >> + >> +/* >> + * Special operations for GDSCs with attached pipe clocks. >> + * The clock should be parked to safe source (tcxo) before turning >> +off the GDSC >> + * and can be switched on as soon as the GDSC is on. >> + * >> + * We remove respective clock sources from clocks map and handle them manually. >> + */ >> +int gdsc_pipe_enable(struct generic_pm_domain *domain) { >> + struct pipe_clk_gdsc *sc = domain_to_pipe_clk_gdsc(domain); >> + int i, ret; >> + >> + ret = gdsc_enable(domain); >> + if (ret) >> + return ret; >> + >> + for (i = 0; i< sc->num_clocks; i++) >> + regmap_update_bits(sc->base.regmap, sc->clocks[i].reg, >> + BIT(sc->clocks[i].shift + sc->clocks[i].width) - BIT(sc->clocks[i].shift), >> + sc->clocks[i].on_value << sc->clocks[i].shift); >> + >> + return 0; >> +} >> +EXPORT_SYMBOL_GPL(gdsc_pipe_enable); >> + >> +int gdsc_pipe_disable(struct generic_pm_domain *domain) { >> + struct pipe_clk_gdsc *sc = domain_to_pipe_clk_gdsc(domain); >> + int i; >> + >> + for (i = sc->num_clocks - 1; i >= 0; i--) >> + regmap_update_bits(sc->base.regmap, sc->clocks[i].reg, >> + BIT(sc->clocks[i].shift + sc->clocks[i].width) - BIT(sc->clocks[i].shift), >> + sc->clocks[i].off_value << sc->clocks[i].shift); >> + >> + /* In case of an error do not try turning the clocks again. We can not be sure about the GDSC state. */ >> + return gdsc_disable(domain); >> +} >> +EXPORT_SYMBOL_GPL(gdsc_pipe_disable); >> diff --git a/drivers/clk/qcom/gdsc.h b/drivers/clk/qcom/gdsc.h index >> d7cc4c21a9d4..b1a2f0abe41c 100644 >> --- a/drivers/clk/qcom/gdsc.h >> +++ b/drivers/clk/qcom/gdsc.h >> @@ -68,11 +68,25 @@ struct gdsc_desc { >> size_t num; >> }; >> >> +struct pipe_clk_gdsc { >> + struct gdsc base; >> + int num_clocks; >> + struct { >> + u32 reg; >> + u32 shift; >> + u32 width; >> + u32 off_value; >> + u32 on_value; >> + } clocks[]; >> +}; >> + >> #ifdef CONFIG_QCOM_GDSC >> int gdsc_register(struct gdsc_desc *desc, struct reset_controller_dev *, >> struct regmap *); >> void gdsc_unregister(struct gdsc_desc *desc); >> int gdsc_gx_do_nothing_enable(struct generic_pm_domain *domain); >> +int gdsc_pipe_enable(struct generic_pm_domain *domain); int >> +gdsc_pipe_disable(struct generic_pm_domain *domain); >> #else >> static inline int gdsc_register(struct gdsc_desc *desc, >> struct reset_controller_dev *rcdev, >> -- >> 2.34.1 >> -- With best wishes Dmitry