Re: [PATCH] w35und: fix usb_control_msg() error handling in wb35_probe()

[Pekka Enberg <penberg@cs.helsinki.fi>]

Sandro Bonazzola wrote:
> Well, I don't know if this can be somehow related to a bug in the driver, but on
> my system, after I've unplugged the USB trasnsitter, ksoftirqd/0 begin to use
> 89% of my CPU time (top). I suspect that there are some issue that should be
> fixed somewhere.

Heh, yes, there's plenty of things to fix. :-)

>> Actually, it's probably me who broke the driver. I think Greg merged
>> most of my patches after 2.6.28 came out. Sandro, if you have some
>> time, please consider doing git bisect to find out which changeset
>> causes the problem.
> 
> Well, actually my 2.6.29 copy seems to be broken:
> arch/x86/kernel/amd_iommu.o: In function `debugfs_create_size_t':
> amd_iommu.c:(.text+0x0): multiple definition of `debugfs_create_size_t'
> arch/x86/kernel/kdebugfs.o:kdebugfs.c:(.text+0x0): first defined here
> make[1]: *** [arch/x86/kernel/built-in.o] Error 1
> make: *** [arch/x86/kernel] Error 2
> 
> # git pull
> Already up-to-date.
> 
> # git status
> # On branch master
> # Changed but not updated:
> #   (use "git add <file>..." to update what will be committed)
> #   (use "git checkout -- <file>..." to discard changes in working directory)
> #
> #       modified:   drivers/staging/winbond/wbusb.c
> #
> no changes added to commit (use "git add" and/or "git commit -a")
> 
> 
> I don't know too much how to use git to do what you're asking. I've some years
> of experience with svn and cvs but git is quite a new tool for me.
> I'll wait a git kernel that compile while I'll try to find out how to do what
> you're asking.

Here's a tutorial on git bisect.

http://kernel.org/pub/software/scm/git/docs/v1.3.3/howto/isolate-bugs-with-bisect.txt

> I have not so much free time to spend in a checkout / build / test if the test
> phase need a kernel reinstall / system reboot. If it's only the driver module I
> think I can try to test it for each changeset in a couple of weeks /starting
> from the day I know how to bisect :-) ).

Assuming 2.6.29-rc1 is broken, you initialize the bisection with:

   git bisect start -- drivers/staging/winbond/
   git bisect good v2.6.28
   git bisect bad v2.6.29-rc1

Then you continue doing this:

   <apply my patch, compile, test>

and

   git bisect good

or

   git bisect bad

depending on whether the kernel worked or not. Repeat until git bisect 
tells you which changeset broke the kernel. The process shouldn't take 
more than 5 compiles/reboots.

		Pekka