>Casper Dik wrote:
>> 
>> Is this not just a case of things changing underneat ipfstat?
>> 
>> Kmem readers do not lock out concurrent activity so they can try to follow
>> stale pointers.
>
>Since I didn't get a response to this, I'll try one more time...
>
>So, yeah, it does seem this only happens when ipf has stuff change under it
>
>BUT, doesn't that imply that when IPF detaches from interfaces its not 
>cleaning up properly? Or am I just not understanding the properly?


No, it means that the kmem readers do not honor locks.

While it follows a foo->next pointer, something in the kernel
holding the appropriate lock can actually do "free(foo->next)"
or even free(foo) and something else may reallocate it the same
instances at which point the kmem readers goes of to never never
land chasing pointers that point no where.

Casper

Reply via email to