| Message ID | 1443050853-24601-1-git-send-email-linus.walleij@linaro.org |
|---|---|
| State | Accepted |
| Commit | f881bab038c9667deab19a85d8666029cbfa6f2c |
| Headers | show |
Hi, On Wed, Sep 23, 2015 at 04:27:33PM -0700, Linus Walleij wrote: > This refactors the changes to the GPIO line naming mechanism to > not have so widespread effects, instead we conclude the patch series > by having created a name attribute in the GPIO descriptor, that need > not be globally unique, and it will be initialized from the old > .names array in struct gpio_chip if it exists, then used in the legacy > sysfs code like the array was used previously. > > The associated changes to name lines from the device tree are > controversial and need to stand alone from this. Resulting changes: > > 1. Remove the export and the header for the gpio_name_to_desc() as so > far the only use is inside gpiolib.c. Staticize gpio_name_to_desc() > and move it above the only function using it. It is used in gpiolib-of.c as well. So I think it should be in drivers/gpio/gpiolib.h. > > 2. Only print a warning if there are two GPIO lines with the same name. > The reason is to preserve current behaviour: before the previous > changes to the naming mechanism this would not reject probing the > driver, instead the error would occur when trying to export the line > in sysfs, so restore this behaviour, but print a friendly warning > if names collide. In gpiolib-of.c is the codepath that adds descriptor names from DT. That uses a warning currently but does not assign the name. Best Regards, Markus > > Cc: Johan Hovold <johan@kernel.org> > Cc: Markus Pargmann <mpa@pengutronix.de> > Signed-off-by: Linus Walleij <linus.walleij@linaro.org> > --- > drivers/gpio/gpiolib.c | 72 +++++++++++++++++++++---------------------- > include/linux/gpio/consumer.h | 1 - > 2 files changed, 35 insertions(+), 38 deletions(-) > > diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c > index 764845c3a4b2..efe8a1072ed0 100644 > --- a/drivers/gpio/gpiolib.c > +++ b/drivers/gpio/gpiolib.c > @@ -90,38 +90,6 @@ struct gpio_desc *gpio_to_desc(unsigned gpio) > EXPORT_SYMBOL_GPL(gpio_to_desc); > > /** > - * Convert a GPIO name to its descriptor > - */ > -struct gpio_desc *gpio_name_to_desc(const char * const name) > -{ > - struct gpio_chip *chip; > - unsigned long flags; > - > - spin_lock_irqsave(&gpio_lock, flags); > - > - list_for_each_entry(chip, &gpio_chips, list) { > - int i; > - > - for (i = 0; i != chip->ngpio; ++i) { > - struct gpio_desc *gpio = &chip->desc[i]; > - > - if (!gpio->name) > - continue; > - > - if (!strcmp(gpio->name, name)) { > - spin_unlock_irqrestore(&gpio_lock, flags); > - return gpio; > - } > - } > - } > - > - spin_unlock_irqrestore(&gpio_lock, flags); > - > - return NULL; > -} > -EXPORT_SYMBOL_GPL(gpio_name_to_desc); > - > -/** > * Get the GPIO descriptor corresponding to the given hw number for this chip. > */ > struct gpio_desc *gpiochip_get_desc(struct gpio_chip *chip, > @@ -250,6 +218,37 @@ static int gpiochip_add_to_list(struct gpio_chip *chip) > return err; > } > > +/** > + * Convert a GPIO name to its descriptor > + */ > +static struct gpio_desc *gpio_name_to_desc(const char * const name) > +{ > + struct gpio_chip *chip; > + unsigned long flags; > + > + spin_lock_irqsave(&gpio_lock, flags); > + > + list_for_each_entry(chip, &gpio_chips, list) { > + int i; > + > + for (i = 0; i != chip->ngpio; ++i) { > + struct gpio_desc *gpio = &chip->desc[i]; > + > + if (!gpio->name) > + continue; > + > + if (!strcmp(gpio->name, name)) { > + spin_unlock_irqrestore(&gpio_lock, flags); > + return gpio; > + } > + } > + } > + > + spin_unlock_irqrestore(&gpio_lock, flags); > + > + return NULL; > +} > + > /* > * Takes the names from gc->names and checks if they are all unique. If they > * are, they are assigned to their gpio descriptors. > @@ -268,11 +267,10 @@ static int gpiochip_set_desc_names(struct gpio_chip *gc) > struct gpio_desc *gpio; > > gpio = gpio_name_to_desc(gc->names[i]); > - if (gpio) { > - dev_err(gc->dev, "Detected name collision for GPIO name '%s'\n", > - gc->names[i]); > - return -EEXIST; > - } > + if (gpio) > + dev_warn(gc->dev, "Detected name collision for " > + "GPIO name '%s'\n", > + gc->names[i]); > } > > /* Then add all names to the GPIO descriptors */ > diff --git a/include/linux/gpio/consumer.h b/include/linux/gpio/consumer.h > index 366a3fdbdbea..a879e3e62379 100644 > --- a/include/linux/gpio/consumer.h > +++ b/include/linux/gpio/consumer.h > @@ -130,7 +130,6 @@ int gpiod_to_irq(const struct gpio_desc *desc); > /* Convert between the old gpio_ and new gpiod_ interfaces */ > struct gpio_desc *gpio_to_desc(unsigned gpio); > int desc_to_gpio(const struct gpio_desc *desc); > -struct gpio_desc *gpio_name_to_desc(const char *name); > > /* Child properties interface */ > struct fwnode_handle; > -- > 2.4.3 > >
On Thu, Sep 24, 2015 at 12:40 AM, Markus Pargmann <mpa@pengutronix.de> wrote: > On Wed, Sep 23, 2015 at 04:27:33PM -0700, Linus Walleij wrote: >> This refactors the changes to the GPIO line naming mechanism to >> not have so widespread effects, instead we conclude the patch series >> by having created a name attribute in the GPIO descriptor, that need >> not be globally unique, and it will be initialized from the old >> .names array in struct gpio_chip if it exists, then used in the legacy >> sysfs code like the array was used previously. >> >> The associated changes to name lines from the device tree are >> controversial and need to stand alone from this. Resulting changes: >> >> 1. Remove the export and the header for the gpio_name_to_desc() as so >> far the only use is inside gpiolib.c. Staticize gpio_name_to_desc() >> and move it above the only function using it. > > It is used in gpiolib-of.c as well. So I think it should be in > drivers/gpio/gpiolib.h. It's not, since I only merged patches 1,2,3,4. So far, all that has changed is that the descriptor has a name property and this is used by sysfs instead of the gc->names[] array. Now the addition of setting names from device tree is controversial and Johan is opposed to it, so we need to discuss more before applying the rest of the patches, and therefore this patch puts in a break after the refactorings (which are anyways very nice). >> 2. Only print a warning if there are two GPIO lines with the same name. >> The reason is to preserve current behaviour: before the previous >> changes to the naming mechanism this would not reject probing the >> driver, instead the error would occur when trying to export the line >> in sysfs, so restore this behaviour, but print a friendly warning >> if names collide. > > In gpiolib-of.c is the codepath that adds descriptor names from DT. That > uses a warning currently but does not assign the name. In later patches. But I do not apply them now. As per above. Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Sep 23, 2015 at 04:27:33PM -0700, Linus Walleij wrote: > This refactors the changes to the GPIO line naming mechanism to > not have so widespread effects, instead we conclude the patch series > by having created a name attribute in the GPIO descriptor, that need > not be globally unique, and it will be initialized from the old > .names array in struct gpio_chip if it exists, then used in the legacy > sysfs code like the array was used previously. > > The associated changes to name lines from the device tree are > controversial and need to stand alone from this. Resulting changes: > > 1. Remove the export and the header for the gpio_name_to_desc() as so > far the only use is inside gpiolib.c. Staticize gpio_name_to_desc() > and move it above the only function using it. > > 2. Only print a warning if there are two GPIO lines with the same name. > The reason is to preserve current behaviour: before the previous > changes to the naming mechanism this would not reject probing the > driver, instead the error would occur when trying to export the line > in sysfs, so restore this behaviour, but print a friendly warning > if names collide. This looks good (apart from the checkpatch warning for the warning message string). You also need to revert ddd5404007b8 ("gpio-sysfs: Use gpio descriptor name instead of gpiochip names array") however as this is an ABI change. Otherwise pins with a name in DT will now be exported using the gpio name rather than number as they used to be. [ The current behaviour is maintained by exporting names from chip->names for hard coded names only. ] Thanks, Johan -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, Oct 4, 2015 at 3:37 PM, Johan Hovold <johan@kernel.org> wrote: > This looks good (apart from the checkpatch warning for the warning > message string). Yeah the GPIO maintainer is liberal about that rule... > You also need to revert ddd5404007b8 ("gpio-sysfs: Use gpio descriptor > name instead of gpiochip names array") however as this is an ABI change. > Otherwise pins with a name in DT will now be exported using the gpio name > rather than number as they used to be. [ The current behaviour is > maintained by exporting names from chip->names for hard coded names > only. ] I think it is ABI-correct: it uses desc->name if that is set, and currently that is only set from chip->names[] so status quo. Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, Oct 04, 2015 at 03:37:42PM +0200, Johan Hovold wrote: > On Wed, Sep 23, 2015 at 04:27:33PM -0700, Linus Walleij wrote: > > This refactors the changes to the GPIO line naming mechanism to > > not have so widespread effects, instead we conclude the patch series > > by having created a name attribute in the GPIO descriptor, that need > > not be globally unique, and it will be initialized from the old > > .names array in struct gpio_chip if it exists, then used in the legacy > > sysfs code like the array was used previously. > > > > The associated changes to name lines from the device tree are > > controversial and need to stand alone from this. Resulting changes: > > > > 1. Remove the export and the header for the gpio_name_to_desc() as so > > far the only use is inside gpiolib.c. Staticize gpio_name_to_desc() > > and move it above the only function using it. > > > > 2. Only print a warning if there are two GPIO lines with the same name. > > The reason is to preserve current behaviour: before the previous > > changes to the naming mechanism this would not reject probing the > > driver, instead the error would occur when trying to export the line > > in sysfs, so restore this behaviour, but print a friendly warning > > if names collide. > > This looks good (apart from the checkpatch warning for the warning > message string). > > You also need to revert ddd5404007b8 ("gpio-sysfs: Use gpio descriptor > name instead of gpiochip names array") however as this is an ABI change. > Otherwise pins with a name in DT will now be exported using the gpio name > rather than number as they used to be. [ The current behaviour is > maintained by exporting names from chip->names for hard coded names > only. ] Even for GPIOs from DT it is not a ABI change. The only GPIOs that have a GPIO name at the moment are using the GPIO hogging mechanism. But hogged GPIOs can't be exported to userspace so there is no difference for these. Best Regards, Markus
On Mon, Oct 05, 2015 at 10:54:52AM +0200, Linus Walleij wrote: > On Sun, Oct 4, 2015 at 3:37 PM, Johan Hovold <johan@kernel.org> wrote: > > > This looks good (apart from the checkpatch warning for the warning > > message string). > > Yeah the GPIO maintainer is liberal about that rule... Heh. I'm sticking to the 80 column limit too, but having grepable error messages do have its merit. > > You also need to revert ddd5404007b8 ("gpio-sysfs: Use gpio descriptor > > name instead of gpiochip names array") however as this is an ABI change. > > Otherwise pins with a name in DT will now be exported using the gpio name > > rather than number as they used to be. [ The current behaviour is > > maintained by exporting names from chip->names for hard coded names > > only. ] > > I think it is ABI-correct: it uses desc->name if that is set, and > currently that is only set from chip->names[] so status quo. Yes, but this work was aiming at generalising the hogs so that they could later be requested (and exported), right? This means that a pin used today from userspace, can not be initialised using this mechanism without breaking the ABI. My suggestion is to leave gpiolib-sysfs.c as is, and make sure to limit the name export to legacy board files as before. Thanks, Johan -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Oct 05, 2015 at 11:47:04AM +0200, Markus Pargmann wrote: > On Sun, Oct 04, 2015 at 03:37:42PM +0200, Johan Hovold wrote: > > You also need to revert ddd5404007b8 ("gpio-sysfs: Use gpio descriptor > > name instead of gpiochip names array") however as this is an ABI change. > > Otherwise pins with a name in DT will now be exported using the gpio name > > rather than number as they used to be. [ The current behaviour is > > maintained by exporting names from chip->names for hard coded names > > only. ] > > Even for GPIOs from DT it is not a ABI change. The only GPIOs that have > a GPIO name at the moment are using the GPIO hogging mechanism. But > hogged GPIOs can't be exported to userspace so there is no difference > for these. Yes, but you're aiming at generalising the hogging mechanism so that such pins can be requested, and that would break the ABI. Let's keep the legacy chip->names and named-gpio exports separate from defining line names in DT for now. Thanks, Johan -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Oct 05, 2015 at 11:07:03AM +0100, Johan Hovold wrote: > On Mon, Oct 05, 2015 at 11:47:04AM +0200, Markus Pargmann wrote: > > On Sun, Oct 04, 2015 at 03:37:42PM +0200, Johan Hovold wrote: > > > > You also need to revert ddd5404007b8 ("gpio-sysfs: Use gpio descriptor > > > name instead of gpiochip names array") however as this is an ABI change. > > > Otherwise pins with a name in DT will now be exported using the gpio name > > > rather than number as they used to be. [ The current behaviour is > > > maintained by exporting names from chip->names for hard coded names > > > only. ] > > > > Even for GPIOs from DT it is not a ABI change. The only GPIOs that have > > a GPIO name at the moment are using the GPIO hogging mechanism. But > > hogged GPIOs can't be exported to userspace so there is no difference > > for these. > > Yes, but you're aiming at generalising the hogging mechanism so that > such pins can be requested, and that would break the ABI. No, hogged GPIOs can not be requested afterwards. But you are right that GPIO names should probably not be added to existing DTs. But that's essentially the same with the currently used GPIO names array. Changing it would change the names in userspace as well. So this would be only useful for all newly created DTs. Best Regards, Markus
On Mon, Oct 5, 2015 at 12:01 PM, Johan Hovold <johan@kernel.org> wrote: >> I think it is ABI-correct: it uses desc->name if that is set, and >> currently that is only set from chip->names[] so status quo. > > Yes, but this work was aiming at generalising the hogs so that they > could later be requested (and exported), right? Yeah something like so... thinking about it we should indeed takes the names directly from chip->names[] for this, so we can use the name field in the gpio descriptor for the "real" chardev interface later. I'll revert the patch. Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Oct 05, 2015 at 12:19:05PM +0200, Markus Pargmann wrote: > On Mon, Oct 05, 2015 at 11:07:03AM +0100, Johan Hovold wrote: > > On Mon, Oct 05, 2015 at 11:47:04AM +0200, Markus Pargmann wrote: > > > On Sun, Oct 04, 2015 at 03:37:42PM +0200, Johan Hovold wrote: > > > > > > You also need to revert ddd5404007b8 ("gpio-sysfs: Use gpio descriptor > > > > name instead of gpiochip names array") however as this is an ABI change. > > > > Otherwise pins with a name in DT will now be exported using the gpio name > > > > rather than number as they used to be. [ The current behaviour is > > > > maintained by exporting names from chip->names for hard coded names > > > > only. ] > > > > > > Even for GPIOs from DT it is not a ABI change. The only GPIOs that have > > > a GPIO name at the moment are using the GPIO hogging mechanism. But > > > hogged GPIOs can't be exported to userspace so there is no difference > > > for these. > > > > Yes, but you're aiming at generalising the hogging mechanism so that > > such pins can be requested, and that would break the ABI. > > No, hogged GPIOs can not be requested afterwards. Ok, you call it something different, but you're adding a way to initialise a gpio during boot that can later be requested from userspace (e.g. "[PATCH v2 3/3] gpiolib: Add GPIO initialization"). My point is that you cannot use this functionality for pins that are currently used from userspace today without breaking the ABI. > But you are right that GPIO names should probably not be added to > existing DTs. But that's essentially the same with the currently used > GPIO names array. Changing it would change the names in userspace as > well. But fortunately the names array is isolated to board files and some legacy SOC controller drivers. We need to continue support that, but we should not make things worse by allowing this to spread to DT. Johan -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Oct 05, 2015 at 01:07:09PM +0200, Linus Walleij wrote: > On Mon, Oct 5, 2015 at 12:01 PM, Johan Hovold <johan@kernel.org> wrote: > > >> I think it is ABI-correct: it uses desc->name if that is set, and > >> currently that is only set from chip->names[] so status quo. > > > > Yes, but this work was aiming at generalising the hogs so that they > > could later be requested (and exported), right? > > Yeah something like so... thinking about it we should indeed > takes the names directly from chip->names[] for this, so we > can use the name field in the gpio descriptor for the "real" > chardev interface later. So what is this "real" chardev interface all about? What are the requirements and features for such an interface? How is it supposed to work? Was this already discussed somewhere? Any chance that you both are at the ELCE? Best Regards, Markus
On Tue, Oct 06, 2015 at 11:13:16AM +0200, Markus Pargmann wrote: > On Mon, Oct 05, 2015 at 01:07:09PM +0200, Linus Walleij wrote: > > On Mon, Oct 5, 2015 at 12:01 PM, Johan Hovold <johan@kernel.org> wrote: > > > > >> I think it is ABI-correct: it uses desc->name if that is set, and > > >> currently that is only set from chip->names[] so status quo. > > > > > > Yes, but this work was aiming at generalising the hogs so that they > > > could later be requested (and exported), right? > > > > Yeah something like so... thinking about it we should indeed > > takes the names directly from chip->names[] for this, so we > > can use the name field in the gpio descriptor for the "real" > > chardev interface later. > > So what is this "real" chardev interface all about? What are the > requirements and features for such an interface? How is it supposed to > work? Was this already discussed somewhere? It has been discussed in various threads, often when someone proposes to tweak the current sysfs interface to cater to their specific needs. Several people have expressed ideas, but no more formal proposal for a new userspace interface has been put forward yet. Since this will also become ABI we need to get it right (this time). > Any chance that you both are at the ELCE? I'm here, but not Linus. I'll see if I can spot you. Johan -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Oct 6, 2015 at 11:13 AM, Markus Pargmann <mpa@pengutronix.de> wrote: > So what is this "real" chardev interface all about? What are the > requirements and features for such an interface? How is it supposed to > work? Was this already discussed somewhere? Comes up all the time. Obviously noone find the time or focus to do this, so I will make a try now to get the ball rolling. Expect a rough chardev patch in a week or so. Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c index 764845c3a4b2..efe8a1072ed0 100644 --- a/drivers/gpio/gpiolib.c +++ b/drivers/gpio/gpiolib.c @@ -90,38 +90,6 @@ struct gpio_desc *gpio_to_desc(unsigned gpio) EXPORT_SYMBOL_GPL(gpio_to_desc); /** - * Convert a GPIO name to its descriptor - */ -struct gpio_desc *gpio_name_to_desc(const char * const name) -{ - struct gpio_chip *chip; - unsigned long flags; - - spin_lock_irqsave(&gpio_lock, flags); - - list_for_each_entry(chip, &gpio_chips, list) { - int i; - - for (i = 0; i != chip->ngpio; ++i) { - struct gpio_desc *gpio = &chip->desc[i]; - - if (!gpio->name) - continue; - - if (!strcmp(gpio->name, name)) { - spin_unlock_irqrestore(&gpio_lock, flags); - return gpio; - } - } - } - - spin_unlock_irqrestore(&gpio_lock, flags); - - return NULL; -} -EXPORT_SYMBOL_GPL(gpio_name_to_desc); - -/** * Get the GPIO descriptor corresponding to the given hw number for this chip. */ struct gpio_desc *gpiochip_get_desc(struct gpio_chip *chip, @@ -250,6 +218,37 @@ static int gpiochip_add_to_list(struct gpio_chip *chip) return err; } +/** + * Convert a GPIO name to its descriptor + */ +static struct gpio_desc *gpio_name_to_desc(const char * const name) +{ + struct gpio_chip *chip; + unsigned long flags; + + spin_lock_irqsave(&gpio_lock, flags); + + list_for_each_entry(chip, &gpio_chips, list) { + int i; + + for (i = 0; i != chip->ngpio; ++i) { + struct gpio_desc *gpio = &chip->desc[i]; + + if (!gpio->name) + continue; + + if (!strcmp(gpio->name, name)) { + spin_unlock_irqrestore(&gpio_lock, flags); + return gpio; + } + } + } + + spin_unlock_irqrestore(&gpio_lock, flags); + + return NULL; +} + /* * Takes the names from gc->names and checks if they are all unique. If they * are, they are assigned to their gpio descriptors. @@ -268,11 +267,10 @@ static int gpiochip_set_desc_names(struct gpio_chip *gc) struct gpio_desc *gpio; gpio = gpio_name_to_desc(gc->names[i]); - if (gpio) { - dev_err(gc->dev, "Detected name collision for GPIO name '%s'\n", - gc->names[i]); - return -EEXIST; - } + if (gpio) + dev_warn(gc->dev, "Detected name collision for " + "GPIO name '%s'\n", + gc->names[i]); } /* Then add all names to the GPIO descriptors */ diff --git a/include/linux/gpio/consumer.h b/include/linux/gpio/consumer.h index 366a3fdbdbea..a879e3e62379 100644 --- a/include/linux/gpio/consumer.h +++ b/include/linux/gpio/consumer.h @@ -130,7 +130,6 @@ int gpiod_to_irq(const struct gpio_desc *desc); /* Convert between the old gpio_ and new gpiod_ interfaces */ struct gpio_desc *gpio_to_desc(unsigned gpio); int desc_to_gpio(const struct gpio_desc *desc); -struct gpio_desc *gpio_name_to_desc(const char *name); /* Child properties interface */ struct fwnode_handle;
This refactors the changes to the GPIO line naming mechanism to not have so widespread effects, instead we conclude the patch series by having created a name attribute in the GPIO descriptor, that need not be globally unique, and it will be initialized from the old .names array in struct gpio_chip if it exists, then used in the legacy sysfs code like the array was used previously. The associated changes to name lines from the device tree are controversial and need to stand alone from this. Resulting changes: 1. Remove the export and the header for the gpio_name_to_desc() as so far the only use is inside gpiolib.c. Staticize gpio_name_to_desc() and move it above the only function using it. 2. Only print a warning if there are two GPIO lines with the same name. The reason is to preserve current behaviour: before the previous changes to the naming mechanism this would not reject probing the driver, instead the error would occur when trying to export the line in sysfs, so restore this behaviour, but print a friendly warning if names collide. Cc: Johan Hovold <johan@kernel.org> Cc: Markus Pargmann <mpa@pengutronix.de> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> --- drivers/gpio/gpiolib.c | 72 +++++++++++++++++++++---------------------- include/linux/gpio/consumer.h | 1 - 2 files changed, 35 insertions(+), 38 deletions(-)