[PATCH] input: gpio-keys - optimize wakeup sequence.

[PATCH] input: gpio-keys - optimize wakeup sequence.

[Abhishek Kumar Singh]

[-- Attachment #1: Type: text/plain, Size: 1193 bytes --]

Dear Mr. Dmitry,

Greetings!



This patch is related to optimization in input key event driver of Kernel module.

Suggested change to avoid the many APIs call chain if there is no key press event triggered.



There is a call back function gpio_keys_resume() called for every suspend/resume of the device.

And whenever this function is called, it is reading the status of the key. 

And gpio_keys_resume() API further calls the below chain of API irrespective of key press event.



APIs call chain:

static void gpio_keys_report_state(struct gpio_keys_drvdata *ddata)

static void gpio_keys_gpio_report_event(struct gpio_button_data *bdata)

gpiod_get_value_cansleep(bdata->gpiod);

input_event(input, type, *bdata->code, state);

input_sync(input);




Suggested changes to avoid the above APIs call chain if there is no key press event triggered.

It will save the device computational resources, power resources and optimize the suspend/resume time"


Please help to review the attached patch and integrate in main line kernel code.

 



Thanks and Regards,
Abhishek Kumar Singh
Sr. Chief Engineer, Samsung Electronics, Noida-India

[-- Attachment #2: input_keys_optimized.zip --]
[-- Type: application/octet-stream, Size: 998 bytes --]

Re: [PATCH] input: gpio-keys - optimize wakeup sequence.

[Gareth Randall]

Dear Mr Singh,

I am not a maintainer but can point out some issues you need to resolve 
with this post.

1. Put the patch in the body of the email and not in an attachment.
2. You need a "Signed-off-by:" line.
3. The email needs to start with a description of the what the patch 
resolves. Don't put "suggested changes".

There are probably other issues as well but I hope this helps you to get 
started. Note that I am not involved in the review process.

Yours,

Gareth

On 17/10/2023 11:34, Abhishek Kumar Singh wrote:
> Dear Mr. Dmitry,
> 
> Greetings!
> 
> 
> 
> This patch is related to optimization in input key event driver of Kernel module.
> 
> Suggested change to avoid the many APIs call chain if there is no key press event triggered.
> 
> 
> 
> There is a call back function gpio_keys_resume() called for every suspend/resume of the device.
> 
> And whenever this function is called, it is reading the status of the key.
> 
> And gpio_keys_resume() API further calls the below chain of API irrespective of key press event.
> 
> 
> 
> APIs call chain:
> 
> static void gpio_keys_report_state(struct gpio_keys_drvdata *ddata)
> 
> static void gpio_keys_gpio_report_event(struct gpio_button_data *bdata)
> 
> gpiod_get_value_cansleep(bdata->gpiod);
> 
> input_event(input, type, *bdata->code, state);
> 
> input_sync(input);
> 
> 
> 
> 
> Suggested changes to avoid the above APIs call chain if there is no key press event triggered.
> 
> It will save the device computational resources, power resources and optimize the suspend/resume time"
> 
> 
> Please help to review the attached patch and integrate in main line kernel code.
> 
>   
> 
> 
> 
> Thanks and Regards,
> Abhishek Kumar Singh
> Sr. Chief Engineer, Samsung Electronics, Noida-India

[PATCH] input: gpio-keys - optimize wakeup sequence.

[Abhishek Kumar Singh]

[-- Attachment #1: Type: text/plain, Size: 2037 bytes --]

Dear Mr. Dmitry,
Greetings!


The patch removes unused many APIs call chain for every suspend/resume of the device 
if no key press event triggered.


There is a call back function gpio_keys_resume() called for
every suspend/resume of the device. and whenever this function called, it is
reading the status of the key. And gpio_keys_resume() API further calls the
below chain of API irrespective of key press event


APIs call chain:
static void gpio_keys_report_state(struct gpio_keys_drvdata *ddata)
static void gpio_keys_gpio_report_event(struct gpio_button_data *bdata)
gpiod_get_value_cansleep(bdata->gpiod);
input_event(input, type, *bdata->code, state);
input_sync(input);


The patch avoid the above APIs call chain if there is no key press event triggered.
It will save the device computational resources, power resources and optimize the suspend/resume time


Signed-off-by: Abhishek Kumar Singh <abhi1.singh@samsung.com>


---
 linux-6.4.11/drivers/input/keyboard/gpio_keys.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)


diff --git a/linux-6.4.11/drivers/input/keyboard/gpio_keys.c b/linux-6.4.11/drivers/input/keyboard/gpio_keys.c
index 13a8ba5..cd1609e 100644
--- a/linux-6.4.11/drivers/input/keyboard/gpio_keys.c
+++ b/linux-6.4.11/drivers/input/keyboard/gpio_keys.c
@@ -687,8 +687,12 @@ static void gpio_keys_report_state(struct gpio_keys_drvdata *ddata)
 
   for (i = 0; i < ddata->pdata->nbuttons; i++) {
     struct gpio_button_data *bdata = &ddata->data[i];
-    if (bdata->gpiod)
-      gpio_keys_gpio_report_event(bdata);
+    if (bdata->gpiod) {
+      struct gpio_keys_button *button = bdata->button;
+      if(button->value) {
+        gpio_keys_gpio_report_event(bdata);
+      }
+    }
   }
   input_sync(input);
 }
---




Thanks and Regards,
Abhishek Kumar Singh
Sr. Chief Engineer, Samsung Electronics, Noida-India
 
  
 


[-- Attachment #2: input_keys_optimized.zip --]
[-- Type: application/octet-stream, Size: 998 bytes --]

Re: [PATCH] input: gpio-keys - optimize wakeup sequence.

[dmitry.torokhov]

Hi Abhishek,

On Thu, Oct 26, 2023 at 11:23:20AM +0530, Abhishek Kumar Singh wrote:
> Dear Mr. Dmitry,
> Greetings!
> 
> 
> The patch removes unused many APIs call chain for every suspend/resume of the device 
> if no key press event triggered.
> 
> 
> There is a call back function gpio_keys_resume() called for
> every suspend/resume of the device. and whenever this function called, it is
> reading the status of the key. And gpio_keys_resume() API further calls the
> below chain of API irrespective of key press event
> 
> 
> APIs call chain:
> static void gpio_keys_report_state(struct gpio_keys_drvdata *ddata)
> static void gpio_keys_gpio_report_event(struct gpio_button_data *bdata)
> gpiod_get_value_cansleep(bdata->gpiod);
> input_event(input, type, *bdata->code, state);
> input_sync(input);
> 
> 
> The patch avoid the above APIs call chain if there is no key press event triggered.
> It will save the device computational resources, power resources and optimize the suspend/resume time

Unfortunately it also breaks the driver as button->value does not hold
the current state of the GPIO but rather set one via device tree so that
the driver can use that value when sending EV_ABS events. So with
typical GPIO-backed keys or buttons you change results in no events
reported on resume.

I also wonder what kind of measurements you did on improvements to
suspend/resume time with your change.

Thanks.

-- 
Dmitry