Re: [PATCH 1/4] ACPI: Add acpi_pr_<level>() interfaces

[Toshi Kani <toshi.kani@hp.com>]

On Thu, 2012-07-19 at 17:43 -0600, Toshi Kani wrote:
> On Thu, 2012-07-19 at 16:32 -0600, Shuah Khan wrote:
> > On Thu, 2012-07-19 at 14:51 -0600, Toshi Kani wrote:
> > 
> > > If your concern is actually a performance bottleneck in acpi_get_name()
> > > you found in the code, you should report it to the ACPI CA team.
> > 
> > I have tried my best to get you to understand the problems in bigger
> > picture your patch set can exacerbate. Looking to somebody else to fix
> > the problems doesn't help. It doesn't look like we can come to an
> > agreement here, we just have to agree to disagree.
> 
> I am not asking someone to fix it.  I tried my best to explain that
> acpi_get_name() does not lead any performance issue when it is called in
> the error paths of ACPI drivers, and why we have to call it to obtain an
> object path info for error analysis.  If you still believe there is a
> performance issue in calling acpi_get_name() under this context, please
> help us understand where the performance bottleneck is in the code.  (I
> hope you just concerned it because it has "acpi_" prefix...) I will then
> work on the issue with the ACPI CA team.

I have measured acpi_pr_<level>() to make sure my statement is correct.
Here are the results:

Avg. acpi_get_name()		 587 ns
Avg. printk()			3420 ns
Avg. kfree()			 238 ns
Avg. acpi_get_time()+kfree()	 825 ns

The results indicate that acpi_pr_<level>() has 20% increase of the time
compared to the regular printk(), which is less than 1 us.  I believe
the results endorse my statement, and may not cause any performance
issue to the error paths of the ACPI drivers.

-Toshi


> Thanks,
> -Toshi
> 
> 
> 
> > caio,
> > -- Shuah
> > 
>