status of kernel memory debugging?

status of kernel memory debugging?

[Willem de Bruijn]

hi all,

For a project in packet filtering (ffpf.sf.net) I'm in the process of 
debugging kernel code I wrote. I have the habit of running all userspace code 
through valgrind, and have setup user-mode-linux to do the same with the 
kernel. 

Reading through some old LKML threads I see that there has been talk of 
valgrinding a UML image, but the outcome appears inconclusive. Could someone 
please update me on the status of memory debugging in the kernel, especially 
regarding valgrind?

Do you have other tools you regularly use, or is is simply a trial-and-error 
practice? Note that I already export my kernelspace code to userspace for 
unit-testing where possible. It is only in the ioctl/syscall handlers that 
this technique fails.

thanks,

  Willem de Bruijn

Re: status of kernel memory debugging?

[Jeff Dike]

On Sat, Jul 30, 2005 at 01:23:27PM +0200, Willem de Bruijn wrote:
> Reading through some old LKML threads I see that there has been talk of 
> valgrinding a UML image, but the outcome appears inconclusive. Could someone 
> please update me on the status of memory debugging in the kernel, especially 
> regarding valgrind?

UML is still too strange for valgrind, despite progress on both sides
(valgrind accepting more strange things and UML becoming less
strange).  

I tried grinding UML a month or so ago, and its use of clone was a
sticking point.

				Jeff

Re: status of kernel memory debugging?

[Willem de Bruijn]

> UML is still too strange for valgrind, despite progress on both sides
> (valgrind accepting more strange things and UML becoming less
> strange).
>
> I tried grinding UML a month or so ago, and its use of clone was a
> sticking point.
>

I think I read your remark, yes. Thanks for the update; I hoped some strides 
had been made in that direction since then. Personally,  I know to little 
about the topic to be of any use.

I guess the best option is then using slab caches per object type, so that I 
can at least find obvious memory leaks. 

On a sidenote, it'll be interesting to see what valgrind reports once (if?) 
the kernel gets a good grinding. 

Willem