| Message ID | 20230725054138.129497-2-quic_fenglinw@quicinc.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v3,1/3] input: pm8xxx-vib: refactor to easily support new SPMI vibrator | expand |
On 25/07/2023 08:16, Fenglin Wu wrote: >>> >>> -static const struct pm8xxx_regs pm8058_regs = { >>> - .drv_addr = 0x4A, >>> - .drv_mask = 0xf8, >>> - .drv_shift = 3, >>> - .drv_en_manual_mask = 0xfc, >>> +static struct reg_field ssbi_vib_regs[VIB_MAX_REG] = { >> >> Change from const to non-const is wrong. How do you support multiple >> devices? No, this is way too fragile now. >> > > The register definition is no longer used as the match data, hw_type is > used. > > The last suggestion was getting the register base address from the DT > and it has to be added into the offset of SPMI vibrator registers > (either in the previous hard-coded format or the later the reg_filed > data structure), so it's not appropriated to make it constant. > > I don't understand this question: "How do you support multiple devices?" > For SSBI vibrator, since all the registers are fixed, and I would assume > that there is no chance to support multiple vibrator devices on the same > SSBI bus. If they are not on the same bus, the regmap device will be > different while the registers definition is the same, and we are still > able to support multiple devices, right? No, you have static memory. One device probes and changes static memory to reg+=base1. Second device probes and changes the same to reg+=base2. > The similar story for SPMI vibrators and it can support multiple devices > if they are located on different SPMI bus, or even if they are on the > same SPMI bus but just having different SID or PID. Sorry, such code cannot go in. These must stay const and you must write driver without any static allocations or singleton-like patterns. Best regards, Krzysztof
On 25/07/2023 12:01, kernel test robot wrote: > Hi Fenglin, > > kernel test robot noticed the following build warnings: > > [auto build test WARNING on dtor-input/next] > [also build test WARNING on dtor-input/for-linus linus/master v6.5-rc3 next-20230725] > [If your patch is applied to the wrong git tree, kindly drop us a note. > And when submitting patch, we suggest to use '--base' as documented in > https://git-scm.com/docs/git-format-patch#_base_tree_information] > > url: https://github.com/intel-lab-lkp/linux/commits/Fenglin-Wu/input-pm8xxx-vib-refactor-to-easily-support-new-SPMI-vibrator/20230725-134504 > base: https://git.kernel.org/pub/scm/linux/kernel/git/dtor/input.git next > patch link: https://lore.kernel.org/r/20230725054138.129497-2-quic_fenglinw%40quicinc.com > patch subject: [PATCH v3 1/3] input: pm8xxx-vib: refactor to easily support new SPMI vibrator > config: x86_64-buildonly-randconfig-r002-20230725 (https://download.01.org/0day-ci/archive/20230725/202307251741.PMtlVAgD-lkp@intel.com/config) > compiler: clang version 15.0.7 (https://github.com/llvm/llvm-project.git 8dfdcc7b7bf66834a761bd8de445840ef68e4d1a) > reproduce: (https://download.01.org/0day-ci/archive/20230725/202307251741.PMtlVAgD-lkp@intel.com/reproduce) > > If you fix the issue in a separate patch/commit (i.e. not just a new version of > the same patch/commit), kindly add following tags > | Reported-by: kernel test robot <lkp@intel.com> > | Closes: https://lore.kernel.org/oe-kbuild-all/202307251741.PMtlVAgD-lkp@intel.com/ > > All warnings (new ones prefixed by >>): > >>> drivers/input/misc/pm8xxx-vibrator.c:190:17: warning: cast to smaller integer type 'enum pm8xxx_vib_type' from 'const void *' [-Wvoid-pointer-to-enum-cast] > vib->hw_type = (enum pm8xxx_vib_type)of_device_get_match_data(dev); > ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > 1 warning generated. > Remember to fix all the warnings. Best regards, Krzysztof
On 7/27/2023 3:07 PM, Krzysztof Kozlowski wrote: > On 25/07/2023 08:16, Fenglin Wu wrote: >>>> >>>> -static const struct pm8xxx_regs pm8058_regs = { >>>> - .drv_addr = 0x4A, >>>> - .drv_mask = 0xf8, >>>> - .drv_shift = 3, >>>> - .drv_en_manual_mask = 0xfc, >>>> +static struct reg_field ssbi_vib_regs[VIB_MAX_REG] = { >>> >>> Change from const to non-const is wrong. How do you support multiple >>> devices? No, this is way too fragile now. >>> >> >> The register definition is no longer used as the match data, hw_type is >> used. >> >> The last suggestion was getting the register base address from the DT >> and it has to be added into the offset of SPMI vibrator registers >> (either in the previous hard-coded format or the later the reg_filed >> data structure), so it's not appropriated to make it constant. >> >> I don't understand this question: "How do you support multiple devices?" >> For SSBI vibrator, since all the registers are fixed, and I would assume >> that there is no chance to support multiple vibrator devices on the same >> SSBI bus. If they are not on the same bus, the regmap device will be >> different while the registers definition is the same, and we are still >> able to support multiple devices, right? > > No, you have static memory. One device probes and changes static memory > to reg+=base1. Second device probes and changes the same to reg+=base2. Thanks, got it. I can update it with following 2 options: 1) keep the register definition in 'reg_filed' data structure and make it constant, copy it to a dynamically allocated memory before adding the 'reg_base' to the '.reg' variable. 2) Define the register offsets as constant data and add the 'reg_base' to the 'reg' while using 'regmap_read()'/'regmap_write()' functions. which one is the preferred way? > >> The similar story for SPMI vibrators and it can support multiple devices >> if they are located on different SPMI bus, or even if they are on the >> same SPMI bus but just having different SID or PID. > > Sorry, such code cannot go in. These must stay const and you must write > driver without any static allocations or singleton-like patterns. > > > Best regards, > Krzysztof >
On 27/07/2023 09:43, Fenglin Wu wrote: > > > On 7/27/2023 3:07 PM, Krzysztof Kozlowski wrote: >> On 25/07/2023 08:16, Fenglin Wu wrote: >>>>> >>>>> -static const struct pm8xxx_regs pm8058_regs = { >>>>> - .drv_addr = 0x4A, >>>>> - .drv_mask = 0xf8, >>>>> - .drv_shift = 3, >>>>> - .drv_en_manual_mask = 0xfc, >>>>> +static struct reg_field ssbi_vib_regs[VIB_MAX_REG] = { >>>> >>>> Change from const to non-const is wrong. How do you support multiple >>>> devices? No, this is way too fragile now. >>>> >>> >>> The register definition is no longer used as the match data, hw_type is >>> used. >>> >>> The last suggestion was getting the register base address from the DT >>> and it has to be added into the offset of SPMI vibrator registers >>> (either in the previous hard-coded format or the later the reg_filed >>> data structure), so it's not appropriated to make it constant. >>> >>> I don't understand this question: "How do you support multiple devices?" >>> For SSBI vibrator, since all the registers are fixed, and I would assume >>> that there is no chance to support multiple vibrator devices on the same >>> SSBI bus. If they are not on the same bus, the regmap device will be >>> different while the registers definition is the same, and we are still >>> able to support multiple devices, right? >> >> No, you have static memory. One device probes and changes static memory >> to reg+=base1. Second device probes and changes the same to reg+=base2. > > Thanks, got it. I can update it with following 2 options: > > 1) keep the register definition in 'reg_filed' data structure and make > it constant, copy it to a dynamically allocated memory before adding the > 'reg_base' to the '.reg' variable. > > 2) Define the register offsets as constant data and add the 'reg_base' > to the 'reg' while using 'regmap_read()'/'regmap_write()' functions. > > which one is the preferred way? Depends on the code. I am not sure if 2 would work with regmap_fields. OTOH, I wonder if the device could just create its own regmap instead of using parents? Then there would be no need of this offset dance. Anyway, adding offset only for some variants seems also not needed. You should add offset to each variant, because each device has this offset. Best regards, Krzysztof
On 7/27/2023 5:22 PM, Krzysztof Kozlowski wrote: > On 27/07/2023 09:43, Fenglin Wu wrote: >> >> >> On 7/27/2023 3:07 PM, Krzysztof Kozlowski wrote: >>> On 25/07/2023 08:16, Fenglin Wu wrote: >>>>>> >>>>>> -static const struct pm8xxx_regs pm8058_regs = { >>>>>> - .drv_addr = 0x4A, >>>>>> - .drv_mask = 0xf8, >>>>>> - .drv_shift = 3, >>>>>> - .drv_en_manual_mask = 0xfc, >>>>>> +static struct reg_field ssbi_vib_regs[VIB_MAX_REG] = { >>>>> >>>>> Change from const to non-const is wrong. How do you support multiple >>>>> devices? No, this is way too fragile now. >>>>> >>>> >>>> The register definition is no longer used as the match data, hw_type is >>>> used. >>>> >>>> The last suggestion was getting the register base address from the DT >>>> and it has to be added into the offset of SPMI vibrator registers >>>> (either in the previous hard-coded format or the later the reg_filed >>>> data structure), so it's not appropriated to make it constant. >>>> >>>> I don't understand this question: "How do you support multiple devices?" >>>> For SSBI vibrator, since all the registers are fixed, and I would assume >>>> that there is no chance to support multiple vibrator devices on the same >>>> SSBI bus. If they are not on the same bus, the regmap device will be >>>> different while the registers definition is the same, and we are still >>>> able to support multiple devices, right? >>> >>> No, you have static memory. One device probes and changes static memory >>> to reg+=base1. Second device probes and changes the same to reg+=base2. >> >> Thanks, got it. I can update it with following 2 options: >> >> 1) keep the register definition in 'reg_filed' data structure and make >> it constant, copy it to a dynamically allocated memory before adding the >> 'reg_base' to the '.reg' variable. >> >> 2) Define the register offsets as constant data and add the 'reg_base' >> to the 'reg' while using 'regmap_read()'/'regmap_write()' functions. >> >> which one is the preferred way? > > Depends on the code. I am not sure if 2 would work with regmap_fields. > OTOH, I wonder if the device could just create its own regmap instead of > using parents? Then there would be no need of this offset dance. > > Anyway, adding offset only for some variants seems also not needed. You > should add offset to each variant, because each device has this offset. > > Best regards, > Krzysztof > Thanks for the suggestion. The Qualcomm SPMI device has to use the 'regmap' from its parent with 16 'reg_bits' and 8 'val_bits' config, the higher 8-bit 'reg_bits' is the peripheral ID (PID) and it could be different in different PMICs even for the same type of HW module, and (PID << 8) is the 'reg_base' here. I assume that you are not in favor of copying the constant data into a dynamic allocated memory, so I will go with option 2. Thanks
diff --git a/drivers/input/misc/pm8xxx-vibrator.c b/drivers/input/misc/pm8xxx-vibrator.c index 04cb87efd799..77bb0018d4e1 100644 --- a/drivers/input/misc/pm8xxx-vibrator.c +++ b/drivers/input/misc/pm8xxx-vibrator.c @@ -12,36 +12,36 @@ #include <linux/regmap.h> #include <linux/slab.h> +#define SSBI_VIB_DRV_EN_MANUAL_MASK 0xfc +#define SSBI_VIB_DRV_LEVEL_MASK 0xf8 +#define SSBI_VIB_DRV_SHIFT 3 +#define SPMI_VIB_DRV_LEVEL_MASK 0xff +#define SPMI_VIB_DRV_SHIFT 0 + #define VIB_MAX_LEVEL_mV (3100) #define VIB_MIN_LEVEL_mV (1200) #define VIB_MAX_LEVELS (VIB_MAX_LEVEL_mV - VIB_MIN_LEVEL_mV) #define MAX_FF_SPEED 0xff -struct pm8xxx_regs { - unsigned int enable_addr; - unsigned int enable_mask; +enum pm8xxx_vib_type { + SSBI_VIB, + SPMI_VIB_GEN1, +}; - unsigned int drv_addr; - unsigned int drv_mask; - unsigned int drv_shift; - unsigned int drv_en_manual_mask; +enum { + VIB_DRV_REG, + VIB_EN_REG, + VIB_MAX_REG, }; -static const struct pm8xxx_regs pm8058_regs = { - .drv_addr = 0x4A, - .drv_mask = 0xf8, - .drv_shift = 3, - .drv_en_manual_mask = 0xfc, +static struct reg_field ssbi_vib_regs[VIB_MAX_REG] = { + REG_FIELD(0xf8, 0, 7), }; -static struct pm8xxx_regs pm8916_regs = { - .enable_addr = 0xc046, - .enable_mask = BIT(7), - .drv_addr = 0xc041, - .drv_mask = 0x1F, - .drv_shift = 0, - .drv_en_manual_mask = 0, +static struct reg_field spmi_vib_gen1_regs[VIB_MAX_REG] = { + REG_FIELD(0x41, 0, 4), + REG_FIELD(0x46, 7, 7), }; /** @@ -58,12 +58,12 @@ static struct pm8xxx_regs pm8916_regs = { struct pm8xxx_vib { struct input_dev *vib_input_dev; struct work_struct work; - struct regmap *regmap; - const struct pm8xxx_regs *regs; + struct regmap_field *r_fields[VIB_MAX_REG]; int speed; int level; bool active; u8 reg_vib_drv; + enum pm8xxx_vib_type hw_type; }; /** @@ -75,22 +75,27 @@ static int pm8xxx_vib_set(struct pm8xxx_vib *vib, bool on) { int rc; unsigned int val = vib->reg_vib_drv; - const struct pm8xxx_regs *regs = vib->regs; + u32 mask = SPMI_VIB_DRV_LEVEL_MASK; + u32 shift = SPMI_VIB_DRV_SHIFT; + + if (vib->hw_type == SSBI_VIB) { + mask = SSBI_VIB_DRV_LEVEL_MASK; + shift = SSBI_VIB_DRV_SHIFT; + } if (on) - val |= (vib->level << regs->drv_shift) & regs->drv_mask; + val |= (vib->level << shift) & mask; else - val &= ~regs->drv_mask; + val &= ~mask; - rc = regmap_write(vib->regmap, regs->drv_addr, val); + rc = regmap_field_write(vib->r_fields[VIB_DRV_REG], val); if (rc < 0) return rc; vib->reg_vib_drv = val; - if (regs->enable_mask) - rc = regmap_update_bits(vib->regmap, regs->enable_addr, - regs->enable_mask, on ? ~0 : 0); + if (vib->hw_type != SSBI_VIB) + rc = regmap_field_write(vib->r_fields[VIB_EN_REG], on); return rc; } @@ -102,13 +107,6 @@ static int pm8xxx_vib_set(struct pm8xxx_vib *vib, bool on) static void pm8xxx_work_handler(struct work_struct *work) { struct pm8xxx_vib *vib = container_of(work, struct pm8xxx_vib, work); - const struct pm8xxx_regs *regs = vib->regs; - int rc; - unsigned int val; - - rc = regmap_read(vib->regmap, regs->drv_addr, &val); - if (rc < 0) - return; /* * pmic vibrator supports voltage ranges from 1.2 to 3.1V, so @@ -168,38 +166,65 @@ static int pm8xxx_vib_probe(struct platform_device *pdev) { struct pm8xxx_vib *vib; struct input_dev *input_dev; - int error; + struct device *dev = &pdev->dev; + struct regmap *regmap; + struct reg_field *regs; + int error, i; unsigned int val; - const struct pm8xxx_regs *regs; + u32 reg_base; - vib = devm_kzalloc(&pdev->dev, sizeof(*vib), GFP_KERNEL); + vib = devm_kzalloc(dev, sizeof(*vib), GFP_KERNEL); if (!vib) return -ENOMEM; - vib->regmap = dev_get_regmap(pdev->dev.parent, NULL); - if (!vib->regmap) + regmap = dev_get_regmap(dev->parent, NULL); + if (!regmap) return -ENODEV; - input_dev = devm_input_allocate_device(&pdev->dev); + input_dev = devm_input_allocate_device(dev); if (!input_dev) return -ENOMEM; INIT_WORK(&vib->work, pm8xxx_work_handler); vib->vib_input_dev = input_dev; - regs = of_device_get_match_data(&pdev->dev); + vib->hw_type = (enum pm8xxx_vib_type)of_device_get_match_data(dev); - /* operate in manual mode */ - error = regmap_read(vib->regmap, regs->drv_addr, &val); + regs = ssbi_vib_regs; + if (vib->hw_type != SSBI_VIB) { + error = fwnode_property_read_u32(dev->fwnode, "reg", ®_base); + if (error < 0) { + dev_err(dev, "Failed to read reg address, rc=%d\n", error); + return error; + } + + if (vib->hw_type == SPMI_VIB_GEN1) + regs = spmi_vib_gen1_regs; + + for (i = 0; i < VIB_MAX_REG; i++) + if (regs[i].reg != 0) + regs[i].reg += reg_base; + } + + error = devm_regmap_field_bulk_alloc(dev, regmap, vib->r_fields, regs, VIB_MAX_REG); if (error < 0) + { + dev_err(dev, "Failed to allocate regmap failed, rc=%d\n", error); return error; + } - val &= regs->drv_en_manual_mask; - error = regmap_write(vib->regmap, regs->drv_addr, val); + error = regmap_field_read(vib->r_fields[VIB_DRV_REG], &val); if (error < 0) return error; - vib->regs = regs; + /* operate in manual mode */ + if (vib->hw_type == SSBI_VIB) { + val &= SSBI_VIB_DRV_EN_MANUAL_MASK; + error = regmap_field_write(vib->r_fields[VIB_DRV_REG], val); + if (error < 0) + return error; + } + vib->reg_vib_drv = val; input_dev->name = "pm8xxx_vib_ffmemless"; @@ -211,14 +236,13 @@ static int pm8xxx_vib_probe(struct platform_device *pdev) error = input_ff_create_memless(input_dev, NULL, pm8xxx_vib_play_effect); if (error) { - dev_err(&pdev->dev, - "couldn't register vibrator as FF device\n"); + dev_err(dev, "couldn't register vibrator as FF device\n"); return error; } error = input_register_device(input_dev); if (error) { - dev_err(&pdev->dev, "couldn't register input device\n"); + dev_err(dev, "couldn't register input device\n"); return error; } @@ -239,9 +263,9 @@ static int pm8xxx_vib_suspend(struct device *dev) static DEFINE_SIMPLE_DEV_PM_OPS(pm8xxx_vib_pm_ops, pm8xxx_vib_suspend, NULL); static const struct of_device_id pm8xxx_vib_id_table[] = { - { .compatible = "qcom,pm8058-vib", .data = &pm8058_regs }, - { .compatible = "qcom,pm8921-vib", .data = &pm8058_regs }, - { .compatible = "qcom,pm8916-vib", .data = &pm8916_regs }, + { .compatible = "qcom,pm8058-vib", .data = (void *)SSBI_VIB }, + { .compatible = "qcom,pm8921-vib", .data = (void *)SSBI_VIB }, + { .compatible = "qcom,pm8916-vib", .data = (void *)SPMI_VIB_GEN1 }, { } }; MODULE_DEVICE_TABLE(of, pm8xxx_vib_id_table);
Currently, all vibrator control register addresses are hard coded, including the base address and the offset, it's not flexible to support new SPMI vibrator module which is usually included in different PMICs with different base address. Refactor this by introducing the HW type terminology and contain the register offsets/masks/shifts in reg_filed data structures corresponding to each vibrator type, and the base address value defined in 'reg' property will be added for SPMI vibrators. Signed-off-by: Fenglin Wu <quic_fenglinw@quicinc.com> --- drivers/input/misc/pm8xxx-vibrator.c | 130 ++++++++++++++++----------- 1 file changed, 77 insertions(+), 53 deletions(-)