Re: 2.5.1-pre5 not easy to boot with devfs

[torvalds@transmeta.com (Linus Torvalds)]

In article <3C085FF3.813BAA57@wanadoo.fr>,
Pierre Rousselet  <pierre.rousselet@wanadoo.fr> wrote:
>As far as I can see,
>
>when CONFIG_DEBUG_KERNEL is set
>  and 
>when devfsd is started at boot time
>I get an Oops when remounting, rw the root fs :
>
>Unable to handle kernel request at va 5a5a5a5e

POISON_BYTE is 0x5a. Something in devfs is using a pointer from a data
structure that was already free'd, and was thus corrupted by poisoning.

(the above is almost certainly just a pointer dereference off 0x5a5a5a5a
with an offset of 4 for some entry at the beginning of a structure,
which is why you get the final "5e" in the page fault address). 

>It boots OK with devfsd when CONFIG_DEBUG_KERNEL is not set.
>It boots OK without devfsd when CONFIG_DEBUG_KERNEL is set (then devfsd
>can be started after login).

Well, not poisoning the free'd memory makes it "work" only in the sense
that usually the free'd memory hasn't been re-allocated yet, so you
don't see the bug even if it is still there.

Richard Gooch probably wants a full stack trace, with symbols. Which
should show it fairly clearly. At least EIP and the first few "stack
trace" entries..

		Linus