Placing ioctl in a kernel module ?

Hi,

Have an application which uses the ioctl call to set configuration of
the wireless card and the IP address. Just wondering if there is some
way this ioctl call can be placed in a kernel module to get better
performance. Any example of code showing the working of ioctl in a
kernel module would be helpful ..

thanks in advance ..
vivekian

Review of any network drivers should produce valuable information for
your purpose(s).

I would recommend looking through the linux/sockios.h header for
specific custom ioctl implementations.

Also it is fairly easy to implement a system wide configuration of
network devices by using the ethtool found at
http://sf.net/projects/gkernel/. Available macros for this utility are
found in linux/ethtool.h.

hth.
Thomas

(E-Mail Removed) wrote:
>
>Have an application which uses the ioctl call to set configuration of
>the wireless card and the IP address. Just wondering if there is some
>way this ioctl call can be placed in a kernel module to get better
>performance.

There is a commonly held misconception that kernel mode is somehow faster
than user mode. It ain't true. The processor runs the same speed in both
modes.

Unless you are starting this process from a kernel driver and spending time
calling out to user-mode, you aren't ging to gain anything.

>Any example of code showing the working of ioctl in a
>kernel module would be helpful ..

There are lots of examples. Look for IoBuildDeviceIoControlRequest --
that's what it's for.
--
- Tim Roberts, (E-Mail Removed)
Providenza & Boekelheide, Inc.

Tim Roberts wrote:
> There is a commonly held misconception that kernel mode is somehow faster
> than user mode. It ain't true. The processor runs the same speed in both
> modes.
>
> Unless you are starting this process from a kernel driver and spending time
> calling out to user-mode, you aren't ging to gain anything.
>

Yes. But , was hoping that could avoid the overhead of userspace to
kernel space transition through this ..

In comp.os.linux.development.system (E-Mail Removed) wrote in part:
>
> Tim Roberts wrote:
>> There is a commonly held misconception that kernel
>> mode is somehow faster than user mode. It ain't true.
>> The processor runs the same speed in both modes.

>> Unless you are starting this process from a kernel driver
>> and spending time calling out to user-mode, you aren't
>> ging to gain anything.

>
> Yes. But , was hoping that could avoid the overhead of
> userspace to kernel space transition through this ..

This time shouldn't be significant. The null syscall (getpid)
takes 295 clocks on my old system. This only matters iff you
have a lot of non-blocking syscalls. Then I'd wonder if it
weren't poor pgm design (polling).

Kernel code isn't any faster, and still has to wait on devices.

One case to hoist code into the kernel may be where you can reduce
the number of interrupts. These can have really slow IO overhead
using the XT-PIC. But using APIC may be a better solution.

-- Robert

Tim Roberts <(E-Mail Removed)> wrote:

>(E-Mail Removed) wrote:
>
>>Any example of code showing the working of ioctl in a
>>kernel module would be helpful ..
>
>There are lots of examples. Look for IoBuildDeviceIoControlRequest --
>that's what it's for.

By the way, I apologize for my momentary disorientation. I read mostly
Windows groups, and IoBuildDeviceIoControlRequest is a Windows kernel
API...
--
- Tim Roberts, (E-Mail Removed)
Providenza & Boekelheide, Inc.