Message ID | 1414825380-15322-1-git-send-email-acourbot@nvidia.com |
---|---|
State | Superseded |
Headers | show |
On Sat, Nov 1, 2014 at 8:03 AM, Alexandre Courbot <acourbot@nvidia.com> wrote: > The issue with this patch is its use of kzalloc() in gpiochip_add(), a > function potentially called during early boot, before kzalloc() becomes > usable. Hence its [RFC] status until we can find a solution for this or > agree that this issue is actually never ran into (from which point can > one use kzalloc()?) So which driver requires that the gpiochip be statically defined... I heard about early call but not *that* early. 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, Nov 3, 2014 at 11:11 PM, Linus Walleij <linus.walleij@linaro.org> wrote: > On Sat, Nov 1, 2014 at 8:03 AM, Alexandre Courbot <acourbot@nvidia.com> wrote: > >> The issue with this patch is its use of kzalloc() in gpiochip_add(), a >> function potentially called during early boot, before kzalloc() becomes >> usable. Hence its [RFC] status until we can find a solution for this or >> agree that this issue is actually never ran into (from which point can >> one use kzalloc()?) > > So which driver requires that the gpiochip be statically defined... > I heard about early call but not *that* early. Let me know if you feel adventurous - in that case I will submit a proper version of this patch to put in -next, and we'll see if people complain... :) -- 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, Nov 4, 2014 at 10:10 AM, Alexandre Courbot <gnurou@gmail.com> wrote: > On Mon, Nov 3, 2014 at 11:11 PM, Linus Walleij <linus.walleij@linaro.org> wrote: >> On Sat, Nov 1, 2014 at 8:03 AM, Alexandre Courbot <acourbot@nvidia.com> wrote: >> >>> The issue with this patch is its use of kzalloc() in gpiochip_add(), a >>> function potentially called during early boot, before kzalloc() becomes >>> usable. Hence its [RFC] status until we can find a solution for this or >>> agree that this issue is actually never ran into (from which point can >>> one use kzalloc()?) >> >> So which driver requires that the gpiochip be statically defined... >> I heard about early call but not *that* early. > > Let me know if you feel adventurous - in that case I will submit a > proper version of this patch to put in -next, and we'll see if people > complain... :) лес рубят — щепки летят as the russians say. Let us have 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
diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c index 50e18a4..013e01e 100644 --- a/drivers/gpio/gpiolib.c +++ b/drivers/gpio/gpiolib.c @@ -47,8 +47,6 @@ */ DEFINE_SPINLOCK(gpio_lock); -static struct gpio_desc gpio_desc[ARCH_NR_GPIOS]; - #define GPIO_OFFSET_VALID(chip, offset) (offset >= 0 && offset < chip->ngpio) static DEFINE_MUTEX(gpio_lookup_lock); @@ -65,10 +63,23 @@ static inline void desc_set_label(struct gpio_desc *d, const char *label) */ struct gpio_desc *gpio_to_desc(unsigned gpio) { - if (WARN(!gpio_is_valid(gpio), "invalid GPIO %d\n", gpio)) - return NULL; - else - return &gpio_desc[gpio]; + struct gpio_chip *chip; + unsigned long flags; + + spin_lock_irqsave(&gpio_lock, flags); + + list_for_each_entry(chip, &gpio_chips, list) { + if (chip->base <= gpio && chip->base + chip->ngpio > gpio) { + spin_unlock_irqrestore(&gpio_lock, flags); + return &chip->desc[gpio - chip->base]; + } + } + + spin_unlock_irqrestore(&gpio_lock, flags); + + pr_warn("invalid GPIO %d\n", gpio); + dump_stack(); + return NULL; } EXPORT_SYMBOL_GPL(gpio_to_desc); @@ -91,7 +102,7 @@ struct gpio_desc *gpiochip_get_desc(struct gpio_chip *chip, */ int desc_to_gpio(const struct gpio_desc *desc) { - return desc - &gpio_desc[0]; + return desc->chip->base + (desc - &desc->chip->desc[0]); } EXPORT_SYMBOL_GPL(desc_to_gpio); @@ -226,12 +237,13 @@ int gpiochip_add(struct gpio_chip *chip) int status = 0; unsigned id; int base = chip->base; + struct gpio_desc *descs; - if ((!gpio_is_valid(base) || !gpio_is_valid(base + chip->ngpio - 1)) - && base >= 0) { - status = -EINVAL; - goto fail; - } + /* TODO context is potentially before kmalloc will work - how can we */ + /* workaround this? */ + descs = kzalloc(sizeof(descs[0] * chip->ngpio, GFP_KERNEL); + if (!descs) + return -ENOMEM; spin_lock_irqsave(&gpio_lock, flags); @@ -247,10 +259,8 @@ int gpiochip_add(struct gpio_chip *chip) status = gpiochip_add_to_list(chip); if (status == 0) { - chip->desc = &gpio_desc[chip->base]; - for (id = 0; id < chip->ngpio; id++) { - struct gpio_desc *desc = &chip->desc[id]; + struct gpio_desc *desc = &descs[id]; desc->chip = chip; /* REVISIT: most hardware initializes GPIOs as @@ -266,6 +276,8 @@ int gpiochip_add(struct gpio_chip *chip) } } + chip->desc = descs; + spin_unlock_irqrestore(&gpio_lock, flags); #ifdef CONFIG_PINCTRL @@ -291,6 +303,9 @@ int gpiochip_add(struct gpio_chip *chip) unlock: spin_unlock_irqrestore(&gpio_lock, flags); fail: + chip->desc = NULL; + kfree(descs); + /* failures here can mean systems won't boot... */ pr_err("%s: GPIOs %d..%d (%s) failed to register\n", __func__, chip->base, chip->base + chip->ngpio - 1, @@ -331,6 +346,9 @@ void gpiochip_remove(struct gpio_chip *chip) list_del(&chip->list); spin_unlock_irqrestore(&gpio_lock, flags); gpiochip_unexport(chip); + + kfree(chip->desc); + chip->descs = NULL; } EXPORT_SYMBOL_GPL(gpiochip_remove);
Replace the ARCH_NR_GPIOS-sized static array of GPIO descriptors by dynamically-allocated arrays for each GPIO chip. This change makes gpio_to_desc() perform in O(n) (where n is the number of GPIO chips registered) instead of O(1), however since n is rarely bigger than 1 or 2 no noticeable performance issue should be expected. Besides this provides more incentive for GPIO consumers to move to the gpiod interface. One could use a O(log(n)) structure to link the GPIO chips together, but considering the low limit of n the hypothetical performance benefit is probably not worth the added complexity. The issue with this patch is its use of kzalloc() in gpiochip_add(), a function potentially called during early boot, before kzalloc() becomes usable. Hence its [RFC] status until we can find a solution for this or agree that this issue is actually never ran into (from which point can one use kzalloc()?) While writing this I noticed a couple of weird things done in gpiochip_add()/gpiochip_remove() and I suspect other people will find them as well, but for now I'd like to focus on the kzalloc() issue and see if we can push this forward. :) Signed-off-by: Alexandre Courbot <acourbot@nvidia.com> --- drivers/gpio/gpiolib.c | 48 +++++++++++++++++++++++++++++++++--------------- 1 file changed, 33 insertions(+), 15 deletions(-)