Re: [v2 3/4] regulator: qcom: Add labibb driver

From: Sumit Semwal
Date: Thu May 14 2020 - 07:47:50 EST


Hello Bjorn,

Thanks very much for the review.

On Tue, 12 May 2020 at 07:46, Bjorn Andersson
<bjorn.andersson@xxxxxxxxxx> wrote:
>
> On Fri 08 May 13:41 PDT 2020, Sumit Semwal wrote:
>
> > From: Nisha Kumari <nishakumari@xxxxxxxxxxxxxx>
> >
> > Qualcomm platforms have LAB(LCD AMOLED Boost)/IBB(Inverting Buck Boost)
> > Regulators, labibb for short, which are used as power supply for
>
> lowercase Regulators
ok.
>
> > LCD Mode displays.
> >
> > This patch adds labibb regulator driver for pmi8998 pmic, found on
>
> Uppercase PMIC
>
ok.
> > SDM845 platforms.
> >
> > Signed-off-by: Nisha Kumari <nishakumari@xxxxxxxxxxxxxx>
> > Signed-off-by: Sumit Semwal <sumit.semwal@xxxxxxxxxx>
> >
> > --
> > v2: sumits: reworked the driver for more common code, and addressed
> > review comments from v1. This includes merging regulator_ops into
> > one, and allowing for future labibb variations.
> > ---
> > drivers/regulator/Kconfig | 10 +
> > drivers/regulator/Makefile | 1 +
> > drivers/regulator/qcom-labibb-regulator.c | 288 ++++++++++++++++++++++
> > 3 files changed, 299 insertions(+)
> > create mode 100644 drivers/regulator/qcom-labibb-regulator.c
> >
> > diff --git a/drivers/regulator/Kconfig b/drivers/regulator/Kconfig
> > index f4b72cb098ef..58704a9fd05d 100644
> > --- a/drivers/regulator/Kconfig
> > +++ b/drivers/regulator/Kconfig
> > @@ -1167,5 +1167,15 @@ config REGULATOR_WM8994
> > This driver provides support for the voltage regulators on the
> > WM8994 CODEC.
> >
> > +config REGULATOR_QCOM_LABIBB
> > + tristate "QCOM LAB/IBB regulator support"
> > + depends on SPMI || COMPILE_TEST
> > + help
> > + This driver supports Qualcomm's LAB/IBB regulators present on the
> > + Qualcomm's PMIC chip pmi8998. QCOM LAB and IBB are SPMI
> > + based PMIC implementations. LAB can be used as positive
> > + boost regulator and IBB can be used as a negative boost regulator
> > + for LCD display panel.
> > +
> > endif
> >
> > diff --git a/drivers/regulator/Makefile b/drivers/regulator/Makefile
> > index 6610ee001d9a..5b313786c0e8 100644
> > --- a/drivers/regulator/Makefile
> > +++ b/drivers/regulator/Makefile
> > @@ -87,6 +87,7 @@ obj-$(CONFIG_REGULATOR_MT6323) += mt6323-regulator.o
> > obj-$(CONFIG_REGULATOR_MT6358) += mt6358-regulator.o
> > obj-$(CONFIG_REGULATOR_MT6380) += mt6380-regulator.o
> > obj-$(CONFIG_REGULATOR_MT6397) += mt6397-regulator.o
> > +obj-$(CONFIG_REGULATOR_QCOM_LABIBB) += qcom-labibb-regulator.o
> > obj-$(CONFIG_REGULATOR_QCOM_RPM) += qcom_rpm-regulator.o
> > obj-$(CONFIG_REGULATOR_QCOM_RPMH) += qcom-rpmh-regulator.o
> > obj-$(CONFIG_REGULATOR_QCOM_SMD_RPM) += qcom_smd-regulator.o
> > diff --git a/drivers/regulator/qcom-labibb-regulator.c b/drivers/regulator/qcom-labibb-regulator.c
> > new file mode 100644
> > index 000000000000..a9dc7c060375
> > --- /dev/null
> > +++ b/drivers/regulator/qcom-labibb-regulator.c
> > @@ -0,0 +1,288 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +// Copyright (c) 2019, The Linux Foundation. All rights reserved.
> > +
> > +#include <linux/module.h>
> > +#include <linux/of_irq.h>
> > +#include <linux/of.h>
> > +#include <linux/of_device.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/regmap.h>
> > +#include <linux/regulator/driver.h>
> > +#include <linux/regulator/of_regulator.h>
> > +
> > +#define REG_PERPH_TYPE 0x04
> > +#define QCOM_LAB_TYPE 0x24
> > +#define QCOM_IBB_TYPE 0x20
> > +
> > +#define REG_LABIBB_STATUS1 0x08
> > +#define REG_LABIBB_ENABLE_CTL 0x46
> > +#define LABIBB_STATUS1_VREG_OK_BIT BIT(7)
> > +#define LABIBB_CONTROL_ENABLE BIT(7)
> > +
> > +#define LAB_ENABLE_CTL_MASK BIT(7)
> > +#define IBB_ENABLE_CTL_MASK (BIT(7) | BIT(6))
> > +
> > +#define POWER_DELAY 8000
> > +
> > +struct labibb_regulator {
> > + struct regulator_desc desc;
> > + struct device *dev;
> > + struct regmap *regmap;
> > + struct regulator_dev *rdev;
> > + u16 base;
> > + u8 type;
> > +};
> > +
> > +struct qcom_labibb {
>
> You pretty much use this as a local variable within probe, and then you
> use labibb_regulator in runtime. Perhaps you can just drop it?
>
Yes, you're right. It's probably a leftover from the re-design. I will drop it.
> > + struct device *dev;
> > + struct regmap *regmap;
> > + struct labibb_regulator lab;
> > + struct labibb_regulator ibb;
> > +};
> > +
> > +struct labibb_regulator_data {
> > + u16 base;
> > + const char *name;
> > + const char *irq_name;
> > + u8 type;
> > +};
> > +
> > +static int qcom_labibb_regulator_is_enabled(struct regulator_dev *rdev)
> > +{
> > + int ret;
> > + u8 val;
> > + struct labibb_regulator *reg = rdev_get_drvdata(rdev);
> > +
> > + ret = regmap_bulk_read(reg->regmap, reg->base +
> > + REG_LABIBB_STATUS1, &val, 1);
> > + if (ret < 0) {
> > + dev_err(reg->dev, "Read register failed ret = %d\n", ret);
> > + return ret;
> > + }
> > +
> > + if (val & LABIBB_STATUS1_VREG_OK_BIT)
> > + return 1;
> > + else
> > + return 0;
>
> return !!(val & LABIBB_STATUS1_VREG_OK_BIT);
>
> > +}
> > +
> > +static int _check_enabled_with_retries(struct regulator_dev *rdev,
> > + int retries, int enabled)
> > +{
> > + int ret;
> > + struct labibb_regulator *reg = rdev_get_drvdata(rdev);
> > +
> > + while (retries--) {
>
> Mark's suggestion of extending _regulator_enable_delay() to support
> polling is_enable() seems reasonable.
>
> The only complication I can see is that code path currently doesn't have
> any expectations of the regulator not being operational at the end -
> this seems to offer that possibility. So some care needs to be taken
> there.
>
I think when I try Mark's suggestion, there will be a need to
communicate an error if we don't get a successful check of status
after the requested number of retries. So yes, I will try and take
care in the next iteration.
>
> But doing that would allow you to use regulator_enable_regmap() as your
> .enable function directly and you can drop
> qcom_labibb_regulator_enable()
>
Agreed.
> > + /* Wait for a small period before checking REG_LABIBB_STATUS1 */
> > + usleep_range(POWER_DELAY, POWER_DELAY + 200);
> > +
> > + ret = qcom_labibb_regulator_is_enabled(rdev);
> > +
> > + if (ret < 0) {
> > + dev_err(reg->dev, "Can't read %s regulator status\n",
> > + reg->desc.name);
> > + return ret;
> > + }
> > +
> > + if (ret == enabled)
> > + return ret;
> > +
> > + }
> > +
> > + return -EINVAL;
> > +}
> > +
> > +static int qcom_labibb_regulator_enable(struct regulator_dev *rdev)
> > +{
> > + int ret, retries = 10;
> > + struct labibb_regulator *reg = rdev_get_drvdata(rdev);
> > +
> > + ret = regulator_enable_regmap(rdev);
> > +
> > + if (ret < 0) {
> > + dev_err(reg->dev, "Write failed: enable %s regulator\n",
> > + reg->desc.name);
> > + return ret;
> > + }
> > +
> > + ret = _check_enabled_with_retries(rdev, retries, 1);
> > + if (ret < 0) {
> > + dev_err(reg->dev, "retries exhausted: enable %s regulator\n",
> > + reg->desc.name);
> > + return ret;
> > + }
> > +
> > + if (ret)
> > + return 0;
>
> This looks weird; check for timeout, check for errors and then return
> success.

Yes, it does look weird - that's because the
_check_enabled_with_retries returns 1 if it matches what you asked to
check (enable or disable via last argument). That should've been coded
differently.

It should go away with the above mentioned rework.
>
> > +
> > +
> > + dev_err(reg->dev, "Can't enable %s\n", reg->desc.name);
> > + return -EINVAL;
> > +}
> > +
> > +static int qcom_labibb_regulator_disable(struct regulator_dev *rdev)
> > +{
> > + int ret, retries = 2;
> > + struct labibb_regulator *reg = rdev_get_drvdata(rdev);
> > +
> > + ret = regulator_disable_regmap(rdev);
> > +
> > + if (ret < 0) {
> > + dev_err(reg->dev, "Write failed: disable %s regulator\n",
> > + reg->desc.name);
> > + return ret;
> > + }
> > +
> > + ret = _check_enabled_with_retries(rdev, retries, 0);
>
> I don't think we care about waiting for the regulator to turning off,
> might be nice to ensure that an off/on cycle really is allowed to take
> its time though. So specifying an desc->off_on_delay of 8200 * 2 (the
> worst case usleep_range() * retries) should take care of this.

Oh, that's neat. Will update.
>
> And then you should be able to just use regulator_disable_regmap() as
> your .disable function.
+1.
>
> > + if (ret < 0) {
> > + dev_err(reg->dev, "retries exhausted: disable %s regulator\n",
> > + reg->desc.name);
> > + return ret;
> > + }
> > +
> > + if (!ret)
> > + return 0;
> > +
> > + dev_err(reg->dev, "Can't disable %s\n", reg->desc.name);
> > + return -EINVAL;
> > +}
> > +
> > +static struct regulator_ops qcom_labibb_ops = {
> > + .enable = qcom_labibb_regulator_enable,
> > + .disable = qcom_labibb_regulator_disable,
> > + .is_enabled = qcom_labibb_regulator_is_enabled,
> > +};
> > +
> > +static int register_labibb_regulator(struct qcom_labibb *labibb,
> > + const struct labibb_regulator_data *reg_data,
> > + struct device_node *of_node)
> > +{
> > + int ret;
> > + struct labibb_regulator *reg;
> > + struct regulator_config cfg = {};
> > +
> > + if (reg_data->type == QCOM_LAB_TYPE) {
> > + reg = &labibb->lab;
> > + reg->desc.enable_mask = LAB_ENABLE_CTL_MASK;
> > + } else {
> > + reg = &labibb->ibb;
> > + reg->desc.enable_mask = IBB_ENABLE_CTL_MASK;
> > + }
> > +
> > + reg->dev = labibb->dev;
> > + reg->base = reg_data->base;
> > + reg->type = reg_data->type;
> > + reg->regmap = labibb->regmap;
> > + reg->desc.enable_reg = reg->base + REG_LABIBB_ENABLE_CTL;
> > + reg->desc.enable_val = LABIBB_CONTROL_ENABLE;
> > + reg->desc.of_match = reg_data->name;
> > + reg->desc.name = reg_data->name;
> > + reg->desc.owner = THIS_MODULE;
> > + reg->desc.type = REGULATOR_VOLTAGE;
> > + reg->desc.ops = &qcom_labibb_ops;
> > +
> > + cfg.dev = labibb->dev;
> > + cfg.driver_data = reg;
> > + cfg.regmap = labibb->regmap;
> > + cfg.of_node = of_node;
> > +
> > + reg->rdev = devm_regulator_register(labibb->dev, &reg->desc,
> > + &cfg);
>
> Do you really need to wrap this line?
No :) - rework leftover. Will correct.
>
> > + if (IS_ERR(reg->rdev)) {
> > + ret = PTR_ERR(reg->rdev);
> > + dev_err(labibb->dev,
> > + "unable to register %s regulator\n", reg_data->name);
> > + return ret;
> > + }
> > + return 0;
>
> return PTR_ERR_OR_ZERO(reg->rdev);
>
> > +}
> > +
> > +static const struct labibb_regulator_data pmi8998_labibb_data[] = {
> > + {0xde00, "lab", "lab-sc-err", QCOM_LAB_TYPE},
>
> Please shorten the interrupt to just "sc-err".
Ok. And actually then I can remove it off from the
labibb_regulotor_data too, since it will be the same for both the
regulators.
>
> > + {0xdc00, "ibb", "ibb-sc-err", QCOM_IBB_TYPE},
> > + { },
> > +};
> > +
> > +static const struct of_device_id qcom_labibb_match[] = {
> > + { .compatible = "qcom,pmi8998-lab-ibb", .data = &pmi8998_labibb_data},
> > + { },
> > +};
> > +MODULE_DEVICE_TABLE(of, qcom_labibb_match);
> > +
> > +static int qcom_labibb_regulator_probe(struct platform_device *pdev)
> > +{
> > + struct qcom_labibb *labibb;
> > + struct device_node *child;
> > + const struct of_device_id *match;
> > + const struct labibb_regulator_data *reg;
> > + u8 type;
> > + int ret;
> > +
> > + labibb = devm_kzalloc(&pdev->dev, sizeof(*labibb), GFP_KERNEL);
> > + if (!labibb)
> > + return -ENOMEM;
> > +
> > + labibb->regmap = dev_get_regmap(pdev->dev.parent, NULL);
> > + if (!labibb->regmap) {
> > + dev_err(&pdev->dev, "Couldn't get parent's regmap\n");
> > + return -ENODEV;
> > + }
> > +
> > + labibb->dev = &pdev->dev;
> > +
> > + match = of_match_device(qcom_labibb_match, &pdev->dev);
> > + if (!match)
> > + return -ENODEV;
> > +
> > + for (reg = match->data; reg->name; reg++) {
> > + child = of_get_child_by_name(pdev->dev.of_node, reg->name);
> > +
> > + /* TODO: This validates if the type of regulator is indeed
> > + * what's mentioned in DT.
> > + * I'm not sure if this is needed, but we'll keep it for now.
> > + */
> > + ret = regmap_bulk_read(labibb->regmap,
> > + reg->base + REG_PERPH_TYPE,
> > + &type, 1);
> > + if (ret < 0) {
> > + dev_err(labibb->dev,
> > + "Peripheral type read failed ret=%d\n",
> > + ret);
> > + return -EINVAL;
> > + }
> > +
> > + if ((type != QCOM_LAB_TYPE) && (type != QCOM_IBB_TYPE)) {
> > + dev_err(labibb->dev,
> > + "qcom_labibb: unknown peripheral type\n");
> > + return -EINVAL;
> > + } else if (type != reg->type) {
> > + dev_err(labibb->dev,
> > + "qcom_labibb: type read %x doesn't match DT %x\n",
> > + type, reg->type);
> > + return -EINVAL;
> > + }
> > +
> > + ret = register_labibb_regulator(labibb, reg, child);
> > + if (ret < 0) {
> > + dev_err(&pdev->dev,
> > + "qcom_labibb: error registering %s regulator: %d\n",
> > + child->full_name, ret);
>
> You already printed in register_labibb_regulator() so no need to print
> again.
Got it.
>
> > + return ret;
> > + }
> > + }
> > +
> > + dev_set_drvdata(&pdev->dev, labibb);
>
> rdev_get_drvdata() returns cfg.driver_data, so you don't need this one.
Got it, and it ties to your other comment about struct qcom_labibb. Will update.
>
> Regards,
> Bjorn
>
> > + return 0;
> > +}
> > +
> > +static struct platform_driver qcom_labibb_regulator_driver = {
> > + .driver = {
> > + .name = "qcom-lab-ibb-regulator",
> > + .of_match_table = qcom_labibb_match,
> > + },
> > + .probe = qcom_labibb_regulator_probe,
> > +};
> > +module_platform_driver(qcom_labibb_regulator_driver);
> > +
> > +MODULE_DESCRIPTION("Qualcomm labibb driver");
> > +MODULE_LICENSE("GPL v2");
> > --
> > 2.26.2
> >

Best,
Sumit.