At 6:18 PM +0200 8/9/02, Xavier HUMBERT wrote:
>Bill Studenmund <wrstuden@netbsd.org>  wrote :
>
>>  i.e. they both are executing
>>  the write() system call at the same time. We don't support that.
>
>In other words, there's a big lock around kernel space, right ?

Yes.  The only real division is the prioritization of tasks.

In NetBSD I would assume that a non-blocking write to a device that 
supports DMA (or PDMA) would buffer the data and return, allowing 
other write system calls while the DMA hardware or PDMA interrupts 
completed the first write.

In OSX, being Mach-based, I would expect that overlapping write calls 
are possible but may be prevented by the BSD layer.  It depends on 
how much Apple/NeXT have modified BSD to exploit Mach.
-- 
The opinions expressed in this message are mine,
not those of Caltech, JPL, NASA, or the US Government.
Henry.B.Hotz@jpl.nasa.gov, or hbhotz@oxy.edu