Weird wasting of time between ioctl() and ioctl dispatcher

Weird wasting of time between ioctl() and ioctl dispatcher

[Cestonaro, Thilo (external)]

Hey,

I'm a developer for Fujitsu Siemens Computers, working on a program which has it's own kernel modules and userland components.
Now cause the program should be released we have done some testing and during this testphase a wierd wasting of time occured
during the call of the ioctl() in the userland component and the actuall entering of the dispatcher function in the module.
It takes 3 min. until the call at last enters my dispatcher. (Debugging output with printf on line before ioctl() and printk as first line in the 
dispatch function points that out). The dispatch function is the ioctl part of the fileoperations struct which defines the module stuff.

The kernel I'm running is the 2.6.16.21-0.8-default of SLED10 (32Bit), but installed on a 64Bit machine.
Can anyone point me to, where I can have a look where the waste of time comes from?

The ioctl itself works fine afterwards, but the whole process is time-critical so the 3 min. do hurt :(.

Ciao Thilo 

Re: Weird wasting of time between ioctl() and ioctl dispatcher

[Arjan van de Ven]

On Mon, 2006-11-27 at 09:24 +0100, Cestonaro, Thilo (external) wrote:
> Hey,
> 
> I'm a developer for Fujitsu Siemens Computers, working on a program which has it's own kernel modules and userland components.
> Now cause the program should be released we have done some testing and during this testphase a wierd wasting of time occured
> during the call of the ioctl() in the userland component and the actuall entering of the dispatcher function in the module.
> It takes 3 min. until the call at last enters my dispatcher. (Debugging output with printf on line before ioctl() and printk as first line in the 
> dispatch function points that out). The dispatch function is the ioctl part of the fileoperations struct which defines the module stuff.
> 
> The kernel I'm running is the 2.6.16.21-0.8-default of SLED10 (32Bit), but installed on a 64Bit machine.
> Can anyone point me to, where I can have a look where the waste of time comes from?
> 
> The ioctl itself works fine afterwards, but the whole process is time-critical so the 3 min. do hurt :(.


Hi,

unfortunately you did not post the source code of at least your kernel
driver, nor did you provide an URL to the source, so it's really hard
for us to help you. Would you mind correcting this oversight?

Greetings,
   Arjan van de Ven
-- 
if you want to mail me at work (you don't), use arjan (at) linux.intel.com
Test the interaction between Linux and your BIOS via http://www.linuxfirmwarekit.org

Re: Weird wasting of time between ioctl() and ioctl dispatcher

[Andi Kleen]

"Cestonaro, Thilo \(external\)"    <Thilo.Cestonaro.external@fujitsu-siemens.com> writes:

> I'm a developer for Fujitsu Siemens Computers, working on a program which has it's own kernel modules and userland components.
> Now cause the program should be released we have done some testing and during this testphase a wierd wasting of time occured
> during the call of the ioctl() in the userland component and the actuall entering of the dispatcher function in the module.
> It takes 3 min. until the call at last enters my dispatcher. 

Something else is holding the big kernel lock for that long. Most likely
it's your own code. Consider using a ->unlocked_ioctl

-Andi