Re: NBD oops in 2.5-bk.

[Lou Langholtz]

Rather than pushing the memset zero hack to NBD to work around this 
problem, how about we fix this problem where it should be fixed - from 
within the blk_init_queue() function. I'm sorry I'm not offerring a real 
patch right here myself for this but I'd rather defer that to the blk_* 
API experts. The basics of such a patch though would be to simply add 
memset(&q->kobj, 0, sizeof(struct kobject)) to the top of the 
blk_init_queue() function. Andrew's been super involved and has done 
wonderful things with the block layer so I really want him to think 
about where and how to best to initialize the new kobj field of 
request_queue or to say why he (or someone else as qualified) believes 
it should be the driver's responsibility to init this field (instead of 
the blk_init_queue function init'ing it). Thanks!