From mboxrd@z Thu Jan 1 00:00:00 1970 From: shawn.guo@linaro.org (Shawn Guo) Date: Tue, 22 Jul 2014 15:42:59 +0800 Subject: [[RFC PATCH]] gpio: gpio-mxc: make sure gpio is input when request IRQ In-Reply-To: <53CE107A.1000000@tqsc.de> References: <1405518664-31313-1-git-send-email-list-09_linux_arm@tqsc.de> <20140722062803.GA21229@dragon> <53CE107A.1000000@tqsc.de> Message-ID: <20140722074257.GC21229@dragon> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org + Grant and devicetree list On Tue, Jul 22, 2014 at 09:19:22AM +0200, Markus Niebel wrote: > Am 22.07.2014 08:28, wrote Shawn Guo: > > On Wed, Jul 16, 2014 at 03:51:04PM +0200, Markus Niebel wrote: > >> From: Markus Niebel > >> > >> When requesting an GPIO irq for imx Soc two things are missing: > >> - there is no check, if the GPIO is already requested > >> - there is no check, if the GPIO is configured as input > >> > >> The first case can lead to reconfiguring the GPIO pin from third > >> party while it is used as IRQ pin, second case will (eventually) > >> prevent IRQ from being seen by SOC becaus the pin is driven by > >> Soc > >> > >> This patch tries to implement (logic taken roughly from gpio-omap) > >> - basic check if gpio already requested > >> - if needed requests the gpio and configures as IN. > >> - if gpio is already requested it is only verified if pin is IN > >> - gpio is locked as irq > >> > >> Tested on a not mainlined i.MX6 based hardware with pin configured > >> by bootloader as OUT HIGH and expecting a low active IRQ. > >> > >> Signed-off-by: Markus Niebel > >> --- > >> drivers/gpio/gpio-mxc.c | 35 +++++++++++++++++++++++++++++++++++ > >> 1 file changed, 35 insertions(+) > >> > >> diff --git a/drivers/gpio/gpio-mxc.c b/drivers/gpio/gpio-mxc.c > >> index db83b3c..4316a38 100644 > >> --- a/drivers/gpio/gpio-mxc.c > >> +++ b/drivers/gpio/gpio-mxc.c > >> @@ -175,6 +175,31 @@ static int gpio_set_irq_type(struct irq_data *d, u32 type) > >> u32 gpio = port->bgc.gc.base + gpio_idx; > >> int edge; > >> void __iomem *reg = port->base; > >> + int ret = 0; > >> + > >> + if (!gpiochip_is_requested(&port->bgc.gc, gpio_idx)) { > >> + char label[32]; > >> + > >> + snprintf(label, 32, "gpio%u-irq", gpio); > >> + ret = gpio_request_one(gpio, GPIOF_DIR_IN, label); > > > > I'm not sure it's correct to call gpio_request_one() from .set_irq_type > > hook. It looks like a API usage violation to me. It should really be > > called from client driver. > > Thats why it is an RFC. I add Linus Walleij to the cc-list. > > Let me describe the problem: > > Currently client drivers have simply interrupts and interrupt-parent > in their bindings, but no interrupt-gpios. Therefore in this case a > client does not know about a dedicated gpio which is to be requested > and configured. Okay. I understand your problem now. This is an issue in case we specify a GPIO to be used as interrupt from device tree. In non-DT world, client driver knows this is an interrupt on GPIO, and therefore takes the responsibility to request and configure the GPIO to work as interrupt. But in DT case, client driver does not know whether it's an IRQ line or GPIO based interrupt. The consequence is that GPIO subsystem, OF subsystem, client driver, none of the three is requesting and configuring the GPIO to be used as interrupt. This is a common issue instead of i.MX specific, and should be addressed globally, I think. Shawn > > > > >> + } else { > >> + val = readl(port->base + GPIO_GDIR); > >> + if (val & BIT(gpio_idx)) > >> + ret = -EINVAL; > > > > It says that the GPIO is requested by someone, but we're not really sure > > if it's the correct one, i.e. the one is requesting set_irq_type. > > > > Yes, but the current situation is even worse (in my eyes): an IRQ can be requested and an > independend party can request and configure the gpio as output ... > > >> + } > >> + > >> + if (ret) { > >> + dev_err(port->bgc.gc.dev, "unable to set gpio_idx %u as IN\n", > >> + gpio_idx); > >> + return ret; > >> + } > > > > Having said that, I'm not sure any above changes is really necessary. > > If any, I would say only gpiochip_is_requested() check makes some sense, > > but we should just fail out if the GPIO hasn't been requested. Nothing > > more can be done in there. > > Going that way as a consequence a reworked device tree binding for gpio irq is needed > (just like in the platform data days) when providing a gpio number and the client has > to request gpio and irq - correct me if I'm wrong. > > Maybe here is something needed with deeper knowledge in the gpio subsystem. > > > > >> + > >> + ret = gpio_lock_as_irq(&port->bgc.gc, gpio_idx); > >> + if (ret) { > >> + dev_err(port->bgc.gc.dev, "unable to lock gpio_idx %u for IRQ\n", > >> + gpio_idx); > >> + return ret; > >> + } > > > > This and the following changes do make sense to me. > > > > Shawn > > > >> > >> port->both_edges &= ~(1 << gpio_idx); > >> switch (type) { > >> @@ -231,6 +256,15 @@ static int gpio_set_irq_type(struct irq_data *d, u32 type) > >> return 0; > >> } > >> > >> +static void gpio_irq_shutdown(struct irq_data *d) > >> +{ > >> + struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); > >> + struct mxc_gpio_port *port = gc->private; > >> + u32 gpio_idx = d->hwirq; > >> + > >> + gpio_unlock_as_irq(&port->bgc.gc, gpio_idx); > >> +} > >> + > >> static void mxc_flip_edge(struct mxc_gpio_port *port, u32 gpio) > >> { > >> void __iomem *reg = port->base; > >> @@ -353,6 +387,7 @@ static void __init mxc_gpio_init_gc(struct mxc_gpio_port *port, int irq_base) > >> ct->chip.irq_mask = irq_gc_mask_clr_bit; > >> ct->chip.irq_unmask = irq_gc_mask_set_bit; > >> ct->chip.irq_set_type = gpio_set_irq_type; > >> + ct->chip.irq_shutdown = gpio_irq_shutdown; > >> ct->chip.irq_set_wake = gpio_set_wake_irq; > >> ct->regs.ack = GPIO_ISR; > >> ct->regs.mask = GPIO_IMR; > >> -- > >> 1.7.9.5 > >> > > > > _______________________________________________ > > linux-arm-kernel mailing list > > linux-arm-kernel at lists.infradead.org > > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel > > >