On Thu, Dec 07, 2023 at 10:33:31PM +0200, Dmitry Baryshkov wrote: > On Thu, 7 Dec 2023 at 19:12, Stephan Gerhold <stephan@xxxxxxxxxxx> wrote: > > On Thu, Dec 07, 2023 at 04:06:56PM +0300, Dmitry Baryshkov wrote: > > > The SPM / SAW2 device also provides a voltage regulator functionality > > > with optional AVS (Adaptive Voltage Scaling) support. The exact register > > > sequence and voltage ranges differs from device to device. > > > > > > Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx> > > > > There is some overlap here with the spmi_saw_set_vdd() functionality in > > qcom_spmi-regulator.c, at least for the SoCs with SPMI PMICs (e.g. > > MSM8974 etc). You don't add support for these in this patch series yet > > but I think it would be good to clarify how we would expect to handle > > those. In other words: > > > > - Would we handle them in qcom_spmi-regulator.c and keep the code in > > the spm.c driver only for the non-SPMI platforms? > > > > - Should we add this in a SSBI regulator driver instead for consistency? > > > > - Or should we move the existing functionality in qcom_spmi-regulator.c > > to here? > > The spmi_saw_set_vdd() is a definite hack and ideally it should be > dropped. It is not possible, though, existing msm8996 DT uses that > spmi/saw regulator to power on the CPU cores. We have to remain > compatible with that hack. > But my intent is to have all other platforms use the spm.c (and > migrate msm8996 at some point too). Thanks, sounds reasonable. > > > > > > --- > > > drivers/soc/qcom/spm.c | 221 ++++++++++++++++++++++++++++++++++++++++- > > > include/soc/qcom/spm.h | 9 ++ > > > 2 files changed, 225 insertions(+), 5 deletions(-) > > > > > > diff --git a/drivers/soc/qcom/spm.c b/drivers/soc/qcom/spm.c > > > index 2f0b1bfe7658..595e2afb2141 100644 > > > --- a/drivers/soc/qcom/spm.c > > > +++ b/drivers/soc/qcom/spm.c > > > [...] > > > @@ -238,6 +273,181 @@ void spm_set_low_power_mode(struct spm_driver_data *drv, > > > spm_register_write_sync(drv, SPM_REG_SPM_CTL, ctl_val); > > > } > > > > > > +static int spm_set_voltage_sel(struct regulator_dev *rdev, unsigned int selector) > > > +{ > > > + struct spm_driver_data *drv = rdev_get_drvdata(rdev); > > > + > > > + drv->volt_sel = selector; > > > + > > > + /* Always do the SAW register writes on the corresponding CPU */ > > > + return smp_call_function_single(drv->reg_cpu, drv->reg_data->set_vdd, drv, true); > > > +} > > > + > > > +static int spm_get_voltage_sel(struct regulator_dev *rdev) > > > +{ > > > + struct spm_driver_data *drv = rdev_get_drvdata(rdev); > > > + > > > + return drv->volt_sel; > > > +} > > > + > > > +static const struct regulator_ops spm_reg_ops = { > > > + .set_voltage_sel = spm_set_voltage_sel, > > > + .get_voltage_sel = spm_get_voltage_sel, > > > + .list_voltage = regulator_list_voltage_linear_range, > > > + .set_voltage_time_sel = regulator_set_voltage_time_sel, > > > +}; > > > + > > > +static void smp_set_vdd_v1_1(void *data) > > > +{ > > > + struct spm_driver_data *drv = data; > > > + unsigned int vctl, data0, data1, avs_ctl, sts; > > > + unsigned int vlevel, volt_sel; > > > + bool avs_enabled; > > > + > > > + volt_sel = drv->volt_sel; > > > + vlevel = volt_sel | 0x80; /* band */ > > > + > > > + avs_ctl = spm_register_read(drv, SPM_REG_AVS_CTL); > > > + vctl = spm_register_read(drv, SPM_REG_VCTL); > > > + data0 = spm_register_read(drv, SPM_REG_PMIC_DATA_0); > > > + data1 = spm_register_read(drv, SPM_REG_PMIC_DATA_1); > > > + > > > + avs_enabled = avs_ctl & SPM_1_1_AVS_CTL_AVS_ENABLED; > > > + > > > + /* If AVS is enabled, switch it off during the voltage change */ > > > + if (avs_enabled) { > > > + avs_ctl &= ~SPM_1_1_AVS_CTL_AVS_ENABLED; > > > + spm_register_write(drv, SPM_REG_AVS_CTL, avs_ctl); > > > + } > > > + > > > + /* Kick the state machine back to idle */ > > > + spm_register_write(drv, SPM_REG_RST, 1); > > > + > > > + vctl = FIELD_SET(vctl, SPM_VCTL_VLVL, vlevel); > > > + data0 = FIELD_SET(data0, SPM_PMIC_DATA_0_VLVL, vlevel); > > > + data1 = FIELD_SET(data1, SPM_PMIC_DATA_1_MIN_VSEL, volt_sel); > > > + data1 = FIELD_SET(data1, SPM_PMIC_DATA_1_MAX_VSEL, volt_sel); > > > + > > > + spm_register_write(drv, SPM_REG_VCTL, vctl); > > > + spm_register_write(drv, SPM_REG_PMIC_DATA_0, data0); > > > + spm_register_write(drv, SPM_REG_PMIC_DATA_1, data1); > > > + > > > + if (read_poll_timeout_atomic(spm_register_read, > > > + sts, sts == vlevel, > > > + 1, 200, false, > > > + drv, SPM_REG_STS1)) { > > > + dev_err_ratelimited(drv->dev, "timeout setting the voltage (%x %x)!\n", sts, vlevel); > > > + goto enable_avs; > > > + } > > > + > > > + if (avs_enabled) { > > > + unsigned int max_avs = volt_sel; > > > + unsigned int min_avs = max(max_avs, 4U) - 4; > > > + > > > + avs_ctl = FIELD_SET(avs_ctl, SPM_AVS_CTL_MIN_VLVL, min_avs); > > > + avs_ctl = FIELD_SET(avs_ctl, SPM_AVS_CTL_MAX_VLVL, max_avs); > > > + spm_register_write(drv, SPM_REG_AVS_CTL, avs_ctl); > > > + } > > > + > > > +enable_avs: > > > + if (avs_enabled) { > > > + avs_ctl |= SPM_1_1_AVS_CTL_AVS_ENABLED; > > > + spm_register_write(drv, SPM_REG_AVS_CTL, avs_ctl); > > > + } > > > +} > > > + > > > +static int spm_get_cpu(struct device *dev) > > > +{ > > > + int cpu; > > > + bool found; > > > + > > > + for_each_possible_cpu(cpu) { > > > + struct device_node *cpu_node, *saw_node; > > > + > > > + cpu_node = of_cpu_device_node_get(cpu); > > > + if (!cpu_node) > > > + continue; > > > + > > > + saw_node = of_parse_phandle(cpu_node, "qcom,saw", 0); > > > + found = (saw_node == dev->of_node); > > > + of_node_put(saw_node); > > > + of_node_put(cpu_node); > > > + > > > + if (found) > > > + return cpu; > > > + } > > > + > > > + /* L2 SPM is not bound to any CPU, tie it to CPU0 */ > > > > Is this necessary? I would kind of expect that it's only important that > > this doesn't happen in parallel on multiple CPUs. The lock in the > > regulator core should already ensure that, though. It's somewhat > > expensive to schedule on other cores, especially if they are currently > > idle and power collapsed. > > If I understand correctly, it is the other way around. From the msm > kernels I see that CPU SPM calls are scheduled to be executed only on > the corresponding CPU/core. For L2 we didn't have the CPU, so to keep > the same code path I selected for them to be executed on CPU0. At this > point I'm not even sure if this cpu0 comes from the downstream tree or > not. > Hm, does 8064/8960 set any voltages through the L2 SPM at all? From a quick look I see only the core SPMs being used as regulators, the L2 SPM seems to be only used for idling. I found a msm_spm_apcs_set_vdd() [1] but: - That one skips the smp_call stuff and calls msm_spm_drv_set_vdd() directly. In other words, there is no scheduling to CPU0. - It seems to be only used in krait-regulator which is I think 8974-specific? If you don't need the regulator on the L2 SPM you could just put an error here rather than the CPU0 fallback. Or alternatively I'd try to avoid the smp_call_function_single() for L2 since I don't see why those should need to be called on a specific CPU. Thanks, Stephan [1]: https://git.codelinaro.org/clo/la/kernel/msm/-/blob/LA.AF.1.2.1-08410-8064.0/arch/arm/mach-msm/spm_devices.c#L257