Re: io_request_lock/queue_lock patch

[Jonathan Lahr <lahr@us.ibm.com>]

> You are now browsing the request list without agreeing on what lock is
> being held -- what happens to drivers assuming that io_request_lock
> protects the list? Boom. For 2.4 we simply cannot afford to muck around
> with this, it's jsut too dangerous. For 2.5 I already completely removed
> the io_request_lock (also helps to catch references to it from drivers).

In this patch, io_request_lock and queue_lock are both acquired in  
generic_unplug_device, so request_fn invocations protect request queue 
integrity.  __make_request acquires queue_lock instead of io_request_lock 
thus protecting queue integrity while allowing greater concurrency.

Nevertheless, I understand your unwillingness to change locking as 
pervasive as io_request_lock.  Such changes would of course involve 
risk.  I am simply trying to improve 2.4 i/o performance, since 2.4
could have a long time left to live.  

> I agree with your SCSI approach, it's the same we took. Low level
> drivers must be responsible for their own locking, the mid layer should
> not pre-grab anything for them. 

Yes, calling out of subsystem scope with locks held can cause problems.

Thanks for your feedback.  

Jonathan 

-- 
Jonathan Lahr
IBM Linux Technology Center
Beaverton, Oregon
lahr@us.ibm.com
503-578-3385