Re: Re: [ntdev] User kernel communication via existing kernel objects

Jamey Kirby wrote:

I tinkered with “Big Hanging DIrect IO” model. I just don’t like
having long-term pending IOCTLs.

I have done an awful lot of capture drivers during my storied Windows
career, and in cases where the hardware can’t handle a new buffer for
every frame, this mechanism has proven to be extremely reliable. It has
a number of advantages. The I/O system handles all of the locking and
reference counting; as long as I don’t complete the ioctls, the buffer
will be good, even if the process implodes. That’s important for DMA,
since there is a third party acting on the buffer.

Now, application guys don’t always find overlapped I/O to be “natural”.
Poor babies. Just write them a DLL that hides the ugly details.


Tim Roberts, xxxxx@probo.com
Providenza & Boekelheide, Inc.