Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753238AbaBOMSh (ORCPT ); Sat, 15 Feb 2014 07:18:37 -0500 Received: from saturn.retrosnub.co.uk ([178.18.118.26]:44900 "EHLO saturn.retrosnub.co.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753118AbaBOMSe (ORCPT ); Sat, 15 Feb 2014 07:18:34 -0500 Message-ID: <52FF5B3A.70806@kernel.org> Date: Sat, 15 Feb 2014 12:19:06 +0000 From: Jonathan Cameron User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.3.0 MIME-Version: 1.0 To: Matt Porter , Grant Likely , Rob Herring , =?ISO-8859-1?Q?Beno=EEt_Cousson?= , Tony Lindgren , Pawel Moll , Mark Rutland , Ian Campbell , Kumar Gala , Thierry Reding CC: Linux IIO List , Linux Kernel Mailing List , Devicetree List , Linux PWM List , Linux OMAP List , Linux ARM Kernel List Subject: Re: [PATCH v3 2/6] iio: pulse: add TI ECAP driver References: <1391626901-31684-1-git-send-email-mporter@linaro.org> <1391626901-31684-3-git-send-email-mporter@linaro.org> In-Reply-To: <1391626901-31684-3-git-send-email-mporter@linaro.org> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 05/02/14 19:01, Matt Porter wrote: > Adds support for capturing PWM signals using the TI ECAP peripheral. > This driver supports triggered buffer capture of pulses on multiple > ECAP instances. In addition, the driver supports configurable polarity > of the signal to be captured. > > Signed-off-by: Matt Porter Ignoring the ABI questions - there are a few bits and bobs inline. Sorry again that it took me so long to take a look at this. Ouch, still far too much unread email in my inbox. > --- > drivers/iio/pulse/Kconfig | 20 ++ > drivers/iio/pulse/Makefile | 6 + > drivers/iio/pulse/tiecap.c | 491 +++++++++++++++++++++++++++++++++++++++++++++ > 3 files changed, 517 insertions(+) > create mode 100644 drivers/iio/pulse/Kconfig > create mode 100644 drivers/iio/pulse/Makefile > create mode 100644 drivers/iio/pulse/tiecap.c > > diff --git a/drivers/iio/pulse/Kconfig b/drivers/iio/pulse/Kconfig > new file mode 100644 > index 0000000..9864d4b > --- /dev/null > +++ b/drivers/iio/pulse/Kconfig > @@ -0,0 +1,20 @@ > +# > +# Pulse Capture Devices > +# > +# When adding new entries keep the list in alphabetical order > + > +menu "Pulse Capture Devices" > + > +config IIO_TIECAP > + tristate "TI ECAP Pulse Capture" > + depends on SOC_AM33XX Looks like it rather more specifically depends on the pwm supplied on that soc. I haven't checked but a dependency on that would seem more logical to me. Pitty it doesn't have a generic interface, because then we could relax the dependency and it would allow you to get a lot more build coverage. > + select IIO_BUFFER > + select IIO_TRIGGERED_BUFFER > + help > + If you say yes here you get support for the TI ECAP peripheral > + in pulse capture mode. > + > + This driver can also be built as a module. If so, the module > + will be called tiecap > + > +endmenu > diff --git a/drivers/iio/pulse/Makefile b/drivers/iio/pulse/Makefile > new file mode 100644 > index 0000000..94d4b00 > --- /dev/null > +++ b/drivers/iio/pulse/Makefile > @@ -0,0 +1,6 @@ > +# > +# Makefile for IIO PWM Capture Devices > +# > + > +# When adding new entries keep the list in alphabetical order > +obj-$(CONFIG_IIO_TIECAP) += tiecap.o > diff --git a/drivers/iio/pulse/tiecap.c b/drivers/iio/pulse/tiecap.c > new file mode 100644 > index 0000000..fd96745 > --- /dev/null > +++ b/drivers/iio/pulse/tiecap.c > @@ -0,0 +1,491 @@ > +/* > + * ECAP IIO pulse capture driver > + * > + * Copyright (C) 2014 Linaro Limited > + * Author: Matt Porter > + * > + * This program is free software; you can redistribute it and/or modify > + * it under the terms of the GNU General Public License as published by > + * the Free Software Foundation; either version 2 of the License, or > + * (at your option) any later version. > + */ > + > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > + > +#include "../../pwm/pwm-tipwmss.h" If we need functions from in there, surely it should be in include/linux/pwm instead of burried in the driver tree? > + > +/* ECAP regs and bits */ Please add prefixes to all of these. i.e. #define ECAP_CAP1 etc. This is to avoid the possibility of a naming clash if these rather short names ever turn in in one of the headers included above. > +#define CAP1 0x08 > +#define CAP2 0x0c > +#define ECCTL1 0x28 > +#define ECCTL1_RUN_FREE BIT(15) > +#define ECCTL1_CAPLDEN BIT(8) > +#define ECCTL1_CAP2POL BIT(2) > +#define ECCTL1_CTRRST1 BIT(1) > +#define ECCTL1_CAP1POL BIT(0) > +#define ECCTL2 0x2a > +#define ECCTL2_SYNCO_SEL_DIS BIT(7) > +#define ECCTL2_TSCTR_FREERUN BIT(4) > +#define ECCTL2_REARM BIT(3) > +#define ECCTL2_STOP_WRAP_2 BIT(1) > +#define ECEINT 0x2c > +#define ECFLG 0x2e > +#define ECCLR 0x30 > +#define ECINT_CTRCMP BIT(7) > +#define ECINT_CTRPRD BIT(6) > +#define ECINT_CTROVF BIT(5) > +#define ECINT_CEVT4 BIT(4) > +#define ECINT_CEVT3 BIT(3) > +#define ECINT_CEVT2 BIT(2) > +#define ECINT_CEVT1 BIT(1) > +#define ECINT_ALL (ECINT_CTRCMP | \ > + ECINT_CTRPRD | \ > + ECINT_CTROVF | \ > + ECINT_CEVT4 | \ > + ECINT_CEVT3 | \ > + ECINT_CEVT2 | \ > + ECINT_CEVT1) > + > +/* ECAP driver flags */ > +#define ECAP_POLARITY_HIGH BIT(1) > +#define ECAP_ENABLED BIT(0) > + > +struct ecap_context { > + u32 cap1; > + u32 cap2; > + u16 ecctl1; > + u16 ecctl2; > + u16 eceint; > +}; > + > +struct ecap_state { > + unsigned long flags; > + unsigned int clk_rate; > + void __iomem *regs; > + u32 *buf; > + struct ecap_context ctx; > +}; > + > +#define dev_to_ecap_state(d) iio_priv(dev_to_iio_dev(d)) > + > +static const struct iio_chan_spec ecap_channels[] = { > + { > + .type = IIO_PULSE, > + .info_mask_separate = > + BIT(IIO_CHAN_INFO_RAW) | BIT(IIO_CHAN_INFO_SCALE), > + .scan_index = 0, > + .scan_type = { > + .sign = 'u', > + .realbits = 32, > + .storagebits = 32, > + .endianness = IIO_LE, > + }, > + }, > + IIO_CHAN_SOFT_TIMESTAMP(1) > +}; > + > +static ssize_t ecap_attr_show(struct device *dev, > + struct device_attribute *attr, char *buf) > +{ > + struct ecap_state *state = dev_to_ecap_state(dev); > + > + return sprintf(buf, "%d\n", > + test_bit(ECAP_POLARITY_HIGH, &state->flags)); > +} > + This function needs a name indicating what attr. There may be only one now, but who knows in future. > +static ssize_t ecap_attr_store(struct device *dev, > + struct device_attribute *attr, > + const char *buf, > + size_t len) > +{ > + int ret; > + bool val; > + struct ecap_state *state = dev_to_ecap_state(dev); > + > + if (test_bit(ECAP_ENABLED, &state->flags)) > + return -EINVAL; > + Not obvious how a boolean value corresponds to a polarity. If we change to the abi I suggested earlier, I'd have this as in_waveformX_inverted. > + ret = strtobool(buf, &val); > + if (ret) > + return ret; > + > + if (val) > + set_bit(ECAP_POLARITY_HIGH, &state->flags); > + else > + clear_bit(ECAP_POLARITY_HIGH, &state->flags); > + > + return len; > +} > + > +static IIO_DEVICE_ATTR(pulse_polarity, S_IRUGO | S_IWUSR, > + ecap_attr_show, ecap_attr_store, 0); > + > +static struct attribute *ecap_attributes[] = { > + &iio_dev_attr_pulse_polarity.dev_attr.attr, > + NULL, > +}; > + > +static struct attribute_group ecap_attribute_group = { > + .attrs = ecap_attributes, > +}; > + > +static int ecap_read_raw(struct iio_dev *idev, > + struct iio_chan_spec const *ch, int *val, > + int *val2, long mask) > +{ > + struct ecap_state *state = iio_priv(idev); > + > + switch (mask) { > + case IIO_CHAN_INFO_RAW: Just don't provide the BIT(IIO_CHAN_INFO_RAW) in the info_mask_separate and there will be no userspace interface. Thus indicating it can't be read better than returning a 'fake' value. > + /* > + * Always return 0 as a pulse width sample > + * is only valid in a triggered condition > + */ > + *val = 0; > + return IIO_VAL_INT; > + case IIO_CHAN_INFO_SCALE: > + *val = 0; > + *val2 = NSEC_PER_SEC / state->clk_rate; > + return IIO_VAL_INT_PLUS_NANO; > + default: > + return -EINVAL; > + } > +} > + > +static const struct iio_info ecap_info = { > + .driver_module = THIS_MODULE, > + .attrs = &ecap_attribute_group, > + .read_raw = &ecap_read_raw, > +}; > + > +static irqreturn_t ecap_trigger_handler(int irq, void *private) > +{ > + struct iio_poll_func *pf = private; > + struct iio_dev *idev = pf->indio_dev; > + struct ecap_state *state = iio_priv(idev); > + > + /* Read pulse counter value */ > + *state->buf = readl(state->regs + CAP2); > + To get more accurate timestamps, you could grab that in the top half handler. See how ad7887 does it for example using the utility function iio_pollfunc_store_time. > + iio_push_to_buffers_with_timestamp(idev, state->buf, iio_get_time_ns()); > + > + iio_trigger_notify_done(idev->trig); > + > + return IRQ_HANDLED; > +}; > + > + > +static const struct iio_trigger_ops iio_interrupt_trigger_ops = { > + .owner = THIS_MODULE, > +}; > + > +static irqreturn_t ecap_interrupt_handler(int irq, void *private) > +{ > + struct iio_dev *idev = private; > + struct ecap_state *state = iio_priv(idev); > + u16 ints; > + > + iio_trigger_poll(idev->trig, 0); > + Do you want to clear the interrupt here, or in the try renable callback for the trigger? That would only be called after the data has been grabbed. > + /* Clear CAP2 interrupt */ > + ints = readw(state->regs + ECFLG); > + if (ints & ECINT_CEVT2) > + writew(ECINT_CEVT2, state->regs + ECCLR); > + else > + dev_warn(&idev->dev, "unhandled interrupt flagged: %04x\n", > + ints); > + > + return IRQ_HANDLED; > +} > + > +static int ecap_buffer_predisable(struct iio_dev *idev) > +{ > + struct ecap_state *state = iio_priv(idev); > + int ret = 0; > + u16 ecctl2; > + > + /* Stop capture */ > + clear_bit(ECAP_ENABLED, &state->flags); > + ecctl2 = readw(state->regs + ECCTL2) & ~ECCTL2_TSCTR_FREERUN; > + writew(ecctl2, state->regs + ECCTL2); > + > + /* Disable and clear all interrupts */ > + writew(0, state->regs + ECEINT); > + writew(ECINT_ALL, state->regs + ECCLR); > + By ordering convention this should be in reverse order of the postenable, hence this predisable call should be first in the function. Not sure it matters in reality but is more 'obviously' correct which is always a good thing. > + ret = iio_triggered_buffer_predisable(idev); > + > + pm_runtime_put_sync(idev->dev.parent); > + > + return ret; > +} > + > +static int ecap_buffer_postenable(struct iio_dev *idev) > +{ > + struct ecap_state *state = iio_priv(idev); > + int ret = 0; > + u16 ecctl1, ecctl2; > + > + pm_runtime_get_sync(idev->dev.parent); > + > + /* Configure pulse polarity */ > + ecctl1 = readw(state->regs + ECCTL1); > + if (test_bit(ECAP_POLARITY_HIGH, &state->flags)) { > + /* CAP1 rising, CAP2 falling */ > + ecctl1 |= ECCTL1_CAP2POL; > + ecctl1 &= ~ECCTL1_CAP1POL; > + } else { > + /* CAP1 falling, CAP2 rising */ > + ecctl1 &= ~ECCTL1_CAP2POL; > + ecctl1 |= ECCTL1_CAP1POL; > + } > + writew(ecctl1, state->regs + ECCTL1); > + > + /* Enable CAP2 interrupt */ > + writew(ECINT_CEVT2, state->regs + ECEINT); > + > + /* Enable capture */ > + ecctl2 = readw(state->regs + ECCTL2); > + ecctl2 |= ECCTL2_TSCTR_FREERUN | ECCTL2_REARM; > + writew(ecctl2, state->regs + ECCTL2); > + set_bit(ECAP_ENABLED, &state->flags); > + > + ret = iio_triggered_buffer_postenable(idev); > + > + return ret; > +} > + > +static const struct iio_buffer_setup_ops ecap_buffer_setup_ops = { > + .postenable = &ecap_buffer_postenable, > + .predisable = &ecap_buffer_predisable, > +}; > + > +static void ecap_init_hw(struct iio_dev *idev) > +{ > + struct ecap_state *state = iio_priv(idev); > + > + clear_bit(ECAP_ENABLED, &state->flags); > + set_bit(ECAP_POLARITY_HIGH, &state->flags); > + > + writew(ECCTL1_RUN_FREE | ECCTL1_CAPLDEN | > + ECCTL1_CAP2POL | ECCTL1_CTRRST1, > + state->regs + ECCTL1); > + > + writew(ECCTL2_SYNCO_SEL_DIS | ECCTL2_STOP_WRAP_2, > + state->regs + ECCTL2); > +} > + > +static const struct of_device_id ecap_of_ids[] = { > + { .compatible = "ti,am33xx-ecap" }, > + { /* sentinel */ }, > +}; > +MODULE_DEVICE_TABLE(of, ecap_of_ids); > + > +static int ecap_probe(struct platform_device *pdev) > +{ > + int irq, ret; > + struct iio_dev *idev; > + struct ecap_state *state; > + struct resource *r; > + struct clk *clk; > + struct iio_trigger *trig; > + u16 status; > + > + idev = devm_iio_device_alloc(&pdev->dev, sizeof(struct ecap_state)); > + if (!idev) > + return -ENOMEM; > + > + state = iio_priv(idev); > + > + clk = devm_clk_get(&pdev->dev, "fck"); > + if (IS_ERR(clk)) { > + dev_err(&pdev->dev, "failed to get clock\n"); > + return PTR_ERR(clk); > + } > + > + state->clk_rate = clk_get_rate(clk); > + if (!state->clk_rate) { > + dev_err(&pdev->dev, "failed to get clock rate\n"); > + return -EINVAL; > + } > + > + platform_set_drvdata(pdev, idev); > + > + idev->dev.parent = &pdev->dev; > + idev->name = dev_name(&pdev->dev); > + idev->modes = INDIO_DIRECT_MODE; > + idev->info = &ecap_info; > + idev->channels = ecap_channels; > + /* One h/w capture and one s/w timestamp channel per instance */ Note that the timestamp may or may not be enabled... Technically it's possible to enable the timestamp and not the channel - that's just rather odd though funnily enough it would give access to the period of the waveform in this case... > + idev->num_channels = ARRAY_SIZE(ecap_channels); > + > + trig = devm_iio_trigger_alloc(&pdev->dev, "%s-dev%d", > + idev->name, idev->id); > + if (!trig) > + return -ENOMEM; > + trig->dev.parent = idev->dev.parent; > + iio_trigger_set_drvdata(trig, idev); > + trig->ops = &iio_interrupt_trigger_ops; > + > + ret = iio_trigger_register(trig); > + if (ret) { > + dev_err(&pdev->dev, "failed to register trigger\n"); > + return ret; > + } > + > + ret = iio_triggered_buffer_setup(idev, NULL, > + &ecap_trigger_handler, > + &ecap_buffer_setup_ops); > + if (ret) > + return ret; > + > + irq = platform_get_irq(pdev, 0); > + if (irq < 0) { > + dev_err(&pdev->dev, "no irq is specified\n"); > + return irq; > + } > + ret = devm_request_irq(&pdev->dev, irq, > + &ecap_interrupt_handler, > + 0, dev_name(&pdev->dev), idev); > + if (ret) { > + dev_err(&pdev->dev, "unable to request irq\n"); > + goto uninit_buffer; > + } > + > + r = platform_get_resource(pdev, IORESOURCE_MEM, 0); > + state->regs = devm_ioremap_resource(&pdev->dev, r); > + if (IS_ERR(state->regs)) { > + dev_err(&pdev->dev, "unable to remap registers\n"); > + ret = PTR_ERR(state->regs); > + goto uninit_buffer; > + }; > + > + ret = iio_device_register(idev); > + if (ret < 0) { > + dev_err(&pdev->dev, "unable to register device\n"); > + goto uninit_buffer; > + } > + Unless I missed something scan_bytes won't contain the correct size until the channels have been enabled from userspace. There would be nothing to stop you allocating 'enough' space here. But then if you do that, you might as well do it directly in the state structure rather than with this separate allocation. > + state->buf = devm_kzalloc(&idev->dev, idev->scan_bytes, GFP_KERNEL); > + if (!state->buf) { > + ret = -ENOMEM; > + goto uninit_buffer; > + } > + > + pm_runtime_enable(&pdev->dev); > + pm_runtime_get_sync(&pdev->dev); > + Does this initialize capture? > + status = pwmss_submodule_state_change(pdev->dev.parent, > + PWMSS_ECAPCLK_EN); > + if (!(status & PWMSS_ECAPCLK_EN_ACK)) { > + dev_err(&pdev->dev, "failed to enable PWMSS config space clock\n"); > + ret = -EINVAL; > + goto pwmss_clk_failure; > + } > + > + ecap_init_hw(idev); > + > + pm_runtime_put_sync(&pdev->dev); > + > + return 0; > + > +pwmss_clk_failure: > + pm_runtime_put_sync(&pdev->dev); > + pm_runtime_disable(&pdev->dev); > + iio_device_unregister(idev); > + > +uninit_buffer: > + iio_triggered_buffer_cleanup(idev); > + > + return ret; > +} > + > +static int ecap_remove(struct platform_device *pdev) > +{ > + struct iio_dev *idev = platform_get_drvdata(pdev); > + > + pm_runtime_get_sync(&pdev->dev); > + > + pwmss_submodule_state_change(pdev->dev.parent, PWMSS_ECAPCLK_STOP_REQ); > + > + pm_runtime_put_sync(&pdev->dev); > + pm_runtime_disable(&pdev->dev); > + > + iio_device_unregister(idev); I'd normally expect the device unregister to be the first element of remove (and the last of probe) given it is responsible for providing the user space interfaces. Can you explain why this doesn't make sense here? > + iio_triggered_buffer_cleanup(idev); > + > + return 0; > +} > + > +#ifdef CONFIG_PM_SLEEP > +static int ecap_suspend(struct device *dev) > +{ > + struct ecap_state *state = dev_to_ecap_state(dev); > + > + pm_runtime_get_sync(dev); > + state->ctx.cap1 = readl(state->regs + CAP1); > + state->ctx.cap2 = readl(state->regs + CAP2); > + state->ctx.eceint = readw(state->regs + ECEINT); > + state->ctx.ecctl1 = readw(state->regs + ECCTL1); > + state->ctx.ecctl2 = readw(state->regs + ECCTL2); > + pm_runtime_put_sync(dev); > + > + /* If capture was active, disable ECAP */ > + if (test_bit(ECAP_ENABLED, &state->flags)) > + pm_runtime_put_sync(dev); > + > + return 0; > +} > + > +static int ecap_resume(struct device *dev) > +{ > + struct ecap_state *state = dev_to_ecap_state(dev); > + > + /* If capture was active, enable ECAP */ > + if (test_bit(ECAP_ENABLED, &state->flags)) > + pm_runtime_get_sync(dev); > + > + pm_runtime_get_sync(dev); > + writel(state->ctx.cap1, state->regs + CAP1); > + writel(state->ctx.cap2, state->regs + CAP2); > + writew(state->ctx.eceint, state->regs + ECEINT); > + writew(state->ctx.ecctl1, state->regs + ECCTL1); > + writew(state->ctx.ecctl2, state->regs + ECCTL2); > + pm_runtime_put_sync(dev); > + > + return 0; > +} > +#endif > + > +static SIMPLE_DEV_PM_OPS(ecap_pm_ops, ecap_suspend, ecap_resume); > + > +static struct platform_driver ecap_iio_driver = { > + .driver = { > + .name = "ecap", > + .owner = THIS_MODULE, > + .of_match_table = of_match_ptr(ecap_of_ids), > + .pm = &ecap_pm_ops, > + }, > + .probe = ecap_probe, > + .remove = ecap_remove, > +}; > + > +module_platform_driver(ecap_iio_driver); > + > +MODULE_DESCRIPTION("ECAP IIO pulse capture driver"); > +MODULE_AUTHOR("Matt Porter "); > +MODULE_LICENSE("GPL"); > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/