cvs commit: src/sys/kern uipc_usrreq.c

[Scott Long]

Nate Lawson wrote:
> On Thu, 10 Jun 2004, Alfred Perlstein wrote:
> 
>>* Robert Watson <rwatson at FreeBSD.org> [040610 14:34] wrote:
>>
>>>  - Sam's version of this change, as with the BSD/OS version, made use of
>>>    both a global lock and per-unpcb locks.  However, in practice, the
>>>    global lock covered all accesses, so I have simplified out the unpcb
>>>    locks in the interest of getting this merged faster (reducing the
>>>    overhead but not sacrificing granularity in most cases).  We will want
>>>    to explore possibilities for improving lock granularity in this code in
>>>    the future.
>>
>>I noticed this in the BSD/os version, it was sort of like...
>>"the global lock covers everything, what's the point of the
>>underlying locks..?"
> 
> 
> In my conversation with the BSD/OS guys, there were often cases where they
> went with a more global lock within a subsystem versus untangling
> re-entrant paths, which would be needed for finer-grained locking.  This
> was true for the CAM approach they used where a single mutex per device
> instance and a middle layer lock protected queue handling.
> 
> -Nate

Yes, but as this approach was described to me, I'm not terribly sure
that it would work in an environment with multiple SIMs.  I appreciate
their approach of 'medium-grain locking' a lot, but I question whether
their approach to CAM and to network locking was really sufficient.

Scott