Re: [PATCH v1] leds: introduce ordered workqueue for leds events instead of system_wq
From: Dmitry Rokosov
Date: Tue Sep 03 2024 - 11:00:28 EST
On Tue, Sep 03, 2024 at 03:01:19PM +0100, Lee Jones wrote:
> On Tue, 20 Aug 2024, Dmitry Rokosov wrote:
>
> > This allows to setup ordered workqueue for leds events. This may be
> > useful, because default 'system_wq' does not guarantee execution order
> > of each work_struct, thus for several brightness update requests (for
> > multiple leds), real brightness switch could be in random order.
> >
> > Yes, for sysfs-based leds we have flush_work() call inside
> > brightness_store() operation, but it's blocking call, so userspace
> > caller can be blocked at a long time, which means leds animation stream
> > can be broken.
> >
> > Ordered workqueue has the same behaviour as system_wq + flush_work(),
> > but all scheduled works are async and userspace caller is not blocked,
> > which it better for userspace animation scheduling.
> >
> > Signed-off-by: Alexey Romanov <avromanov@xxxxxxxxxxxxxxxxx>
> > Signed-off-by: Dmitry Rokosov <ddrokosov@xxxxxxxxxxxxxxxxx>
> > ---
> > drivers/leds/led-class.c | 12 +++++++++++-
> > drivers/leds/led-core.c | 6 +++---
> > include/linux/leds.h | 1 +
> > 3 files changed, 15 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/leds/led-class.c b/drivers/leds/led-class.c
> > index ba1be15cfd8e..fba12471cf1f 100644
> > --- a/drivers/leds/led-class.c
> > +++ b/drivers/leds/led-class.c
> > @@ -25,6 +25,8 @@
> > static DEFINE_MUTEX(leds_lookup_lock);
> > static LIST_HEAD(leds_lookup_list);
> >
> > +static struct workqueue_struct *leds_wq;
>
> Does this _have_ to be global?
>
Unfortunately, yes, it has. Each LED described in the DTS, for example,
is represented as led_cdev. We do not have a led_device or a similar
structure to represent the entire LED device (the controller that
integrates all LEDs and manages them).
One option is to move leds_wq to the driver’s private data. However,
from my perspective, this is a more complicated solution, especially
considering that we have the ability to create a global leds_wq variable
in the core implementation, which can resolve the described issue
entirely.
> Isn't there a suitable data structure that we can store it in?
>
> > static ssize_t brightness_show(struct device *dev,
> > struct device_attribute *attr, char *buf)
> > {
> > @@ -57,7 +59,6 @@ static ssize_t brightness_store(struct device *dev,
> > if (state == LED_OFF)
> > led_trigger_remove(led_cdev);
> > led_set_brightness(led_cdev, state);
> > - flush_work(&led_cdev->set_brightness_work);
> >
> > ret = size;
> > unlock:
> > @@ -549,6 +550,8 @@ int led_classdev_register_ext(struct device *parent,
> >
> > led_update_brightness(led_cdev);
> >
> > + led_cdev->wq = leds_wq;
> > +
> > led_init_core(led_cdev);
> >
> > #ifdef CONFIG_LEDS_TRIGGERS
> > @@ -667,12 +670,19 @@ EXPORT_SYMBOL_GPL(devm_led_classdev_unregister);
> >
> > static int __init leds_init(void)
> > {
> > + leds_wq = alloc_ordered_workqueue("leds", 0);
> > + if (!leds_wq) {
> > + pr_err("failed to create leds ordered workqueue\n");
>
> Nit: "LEDs"
>
Okay, I will send v2 patch today with fixed this line.
> > + return -ENOMEM;
> > + }
> > +
> > return class_register(&leds_class);
> > }
> >
> > static void __exit leds_exit(void)
> > {
> > class_unregister(&leds_class);
> > + destroy_workqueue(leds_wq);
> > }
> >
> > subsys_initcall(leds_init);
> > diff --git a/drivers/leds/led-core.c b/drivers/leds/led-core.c
> > index 89c9806cc97f..9769ac49be20 100644
> > --- a/drivers/leds/led-core.c
> > +++ b/drivers/leds/led-core.c
> > @@ -266,7 +266,7 @@ void led_blink_set_nosleep(struct led_classdev *led_cdev, unsigned long delay_on
> > led_cdev->delayed_delay_on = delay_on;
> > led_cdev->delayed_delay_off = delay_off;
> > set_bit(LED_SET_BLINK, &led_cdev->work_flags);
> > - schedule_work(&led_cdev->set_brightness_work);
> > + queue_work(led_cdev->wq, &led_cdev->set_brightness_work);
> > return;
> > }
> >
> > @@ -297,7 +297,7 @@ void led_set_brightness(struct led_classdev *led_cdev, unsigned int brightness)
> > */
> > if (!brightness) {
> > set_bit(LED_BLINK_DISABLE, &led_cdev->work_flags);
> > - schedule_work(&led_cdev->set_brightness_work);
> > + queue_work(led_cdev->wq, &led_cdev->set_brightness_work);
> > } else {
> > set_bit(LED_BLINK_BRIGHTNESS_CHANGE,
> > &led_cdev->work_flags);
> > @@ -333,7 +333,7 @@ void led_set_brightness_nopm(struct led_classdev *led_cdev, unsigned int value)
> > set_bit(LED_SET_BRIGHTNESS_OFF, &led_cdev->work_flags);
> > }
> >
> > - schedule_work(&led_cdev->set_brightness_work);
> > + queue_work(led_cdev->wq, &led_cdev->set_brightness_work);
> > }
> > EXPORT_SYMBOL_GPL(led_set_brightness_nopm);
> >
> > diff --git a/include/linux/leds.h b/include/linux/leds.h
> > index 6300313c46b7..7c9f1cb12ab9 100644
> > --- a/include/linux/leds.h
> > +++ b/include/linux/leds.h
> > @@ -169,6 +169,7 @@ struct led_classdev {
> > int new_blink_brightness;
> > void (*flash_resume)(struct led_classdev *led_cdev);
> >
> > + struct workqueue_struct *wq; /* LED workqueue */
> > struct work_struct set_brightness_work;
> > int delayed_set_value;
> > unsigned long delayed_delay_on;
> > --
> > 2.43.0
> >
> >
>
> --
> Lee Jones [李琼斯]
--
Thank you,
Dmitry