On Fri, Nov 18, 2005 at 10:22:50AM +0900, YAMAMOTO Takashi wrote:
> 
> nfs VOP_GETATTR sometimes attempts to grab pages.  it causes deadlock
> if you already holds the pages.
> 

Now that I think on it, this may have been the reason my test process
was in the D state - I was running the code via a nfs share mounted
over the loopback to try out the page modification detection.

> 
> when pages are read-in is under the control of each filesystems.
> you can't hook it without dealing with filesystem internals.
> 

Ugh - I really don't want to do that.  Last I looked I could not find
where a process gets woken up when the pages are ready, if that exists
(I think it should) then maybe I can put something somewhere (I begin
hand waving) that indicates the process is sleeping on an async page
in of _these_ pages and verify them prior to the process being woken
up.

-- 
Brett Lymn