Re: [linux-iscsi-devel] 5.x branch mainline kernel push in two weeks?

[Mike C. <mic...@cs...>]

Dmitry Yusupov wrote:
> On Mon, 2005-04-18 at 20:47 +0200, Christoph Hellwig wrote:
> 
>>Another "issue" is that there's lots of networking black magic in
>>iscsi_tcp.c.
> 
> 
> and one more "issue" i would like to resolve before next submission -
> using dedicated workqueue(s) instead of generic per-cpu workqueues.
> in case of OOM, per-cpu workqueue could be blocked by other kernel
> thread for a long time. this is potentially could lead to the deadlock
> scenario, especially in case when iSCSI disk is used for swapping.
> 

When you use dm-multipath you have another problem where it uses the 
generic workqueue to resumbit IO and that can sleep. If there is only 
one thread for the workwueue then IO cannot get sent. Maybe dm-multipath 
should be using its workqueue thought too.