[PATCH]optimize error handling in tick_check_new_device

[PATCH]optimize error handling in tick_check_new_device

[Gaurav Jindal (Gaurav Jindal)]

In the function tick_check_new_device, module error is handled after checking
if the newdev should go as broadcast-installation device in the functions
tick_check_percpu and tick_check_preferred. In case the try_module_get returns
false, the above 2 calls are extra thus wasting CPU cycles.

Shifting the try_module_get as first error handling saves CPU cycles when there
is nothing to do in case of module related error.

Signed-off-by:Gaurav Jindal<gaurav.jindal@spreadtrum.com>
Co-developed-by:Sanjeev Yadav<sanjeev.yadav@spreadtrum.com>

---

diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c
index 4fcd99e..23faf4a 100644
--- a/kernel/time/tick-common.c
+++ b/kernel/time/tick-common.c
@@ -307,6 +307,9 @@ void tick_check_new_device(struct clock_event_device *newdev)
 	td = &per_cpu(tick_cpu_device, cpu);
 	curdev = td->evtdev;

+	if (!try_module_get(newdev->owner))
+		return;
+
 	/* cpu local device ? */
 	if (!tick_check_percpu(curdev, newdev, cpu))
 		goto out_bc;
@@ -315,8 +318,6 @@ void tick_check_new_device(struct clock_event_device *newdev)
 	if (!tick_check_preferred(curdev, newdev))
 		goto out_bc;

-	if (!try_module_get(newdev->owner))
-		return;

 	/*
 	 * Replace the eventually existing device by the new

Re: [PATCH]optimize error handling in tick_check_new_device

[Nicolai Stange]

"Gaurav Jindal (Gaurav Jindal)" <Gaurav.Jindal@spreadtrum.com> writes:
> diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c
> index 4fcd99e..23faf4a 100644
> --- a/kernel/time/tick-common.c
> +++ b/kernel/time/tick-common.c
> @@ -307,6 +307,9 @@ void tick_check_new_device(struct clock_event_device *newdev)
>  	td = &per_cpu(tick_cpu_device, cpu);
>  	curdev = td->evtdev;


>
> +	if (!try_module_get(newdev->owner))
> +		return;
> +

Doesn't this leak a reference to newdev->owner if one of the following
conditions hold?

>  	/* cpu local device ? */
>  	if (!tick_check_percpu(curdev, newdev, cpu))
>  		goto out_bc;
> @@ -315,8 +318,6 @@ void tick_check_new_device(struct clock_event_device *newdev)
>  	if (!tick_check_preferred(curdev, newdev))
>  		goto out_bc;



> -	if (!try_module_get(newdev->owner))
> -		return;
>

Thanks,

Nicolai

Re: [PATCH]optimize error handling in tick_check_new_device

[Thomas Gleixner]

On Wed, 27 Jul 2016, Gaurav Jindal (Gaurav Jindal) wrote:
> In the function tick_check_new_device, module error is handled after checking
> if the newdev should go as broadcast-installation device in the functions
> tick_check_percpu and tick_check_preferred. In case the try_module_get returns
> false, the above 2 calls are extra thus wasting CPU cycles.

This is hardly a fast path so the extra cycles do not matter at all. As a
result of your change you leak a module reference when the 'goto out_bc' is
taken. So I prefer the existing correct version over a pretty uninteresting
speedup.

Thanks,

	tglx