> It's still a SunOS style giant kernel lock implementation, we wanted
> to get something working quickly as a proof of concept.
> 
> Next we will be sprinkling spin locks around major subsections of the
> kernel to allow real parallel kernel activity then perhaps an even
> finer grained implementation after that.  (never a true
> per-every-data-structure locking style like Solaris, SGI got SMP right
> and they do medium grained locking, solaris spends 50% of it's time
> playing with locks for only a slight gain in very specialized cases).
> 
> Actually we have a scheme that will make Linux scalable to n-CPU's
> where n can't fit in an unsigned char. ;-)

This all sounds great. The only problem as a NetBSD developer is the
status of the code. Would the authors be willing to let the relevant
mechanisms (that would need a *lot* of porting) be put under BSD style
Copyrights as well as Copylefted ? This may make things easier quicker
for the NetBSD community to integrate changes. It is only slightly more
work to read the code then reimplement, but if it is alreay done...

I susoect that the kernel models are such that only a very small proportion
of the code could be used, and we can more likely just use it as a reference
doc.

Meanwhile, have any of you Linux/Sparc developers got docs for the MACE
ethernet chips used in the Quad Sbus Ethernet card or the chip set
used on the 100BaseTX card ? I have the hardware, but *repeated* calls
and e-mails to the guy at Sun (Tom Giles ?) have been rudely ignored. Not
even a "No, not me - try XXX". sigh.

Regards,
-- 
Peter Galbavy                                           peter@wonderland.org
@ Home                                                 phone://44/973/499465
in Wonderland                              http://www.wonderland.org/~peter/
                               snail://UK/NW1_6LE/London/21_Harewood_Avenue/