> From: Jay Jaeger
> SACK turnaround capability so that the machine doesn't hang accessing
> an address that doesn't respond on the UNIBUS.
Umm, I think you're mixing up i) grant timeouts and ii) master-slave
timeouts.
All PDP-11 CPUs have master-slave timeout handling; after a short delay
(10usec or so) with no SSYN (UNIBUS) or BRPLY (QBUS), they resume processing,
and take an immediate trap. Most OS's (UNIX, for sure) use this when they are
sizing memory.
Grant timeouts are less well-documented. I think most CPUs deal with this (not
receiving a SACK 'fairly quickly' in response to a grant); I think they
basically just ignore t, and keep processing. (That is because there are
legitimate causes for not having a grant ackknowledged; e.g. if a device is
requesting an interrupt, and then just as the CPU sends out a grant, the
device is reset, the device won't respond to the grant, since it's no longer
trying to do an interrupt.)
The 'SACK turnaround' I think is only used with system health verification;
the system wants to make sure that the grant lines aren't broken anywhere.
(That's because _if_ a grant line is broken, devices downstream of the break
can no longer do interrupts, which generally _will_ hang the overall system,
when interrupts don't work as usual.) To do this, the CPU sends an
_un-requested_ grant out (on startup), and the SACK turnaround circuitry on
the terminator turns it around and sends it back to the CPU; when the CPU sees
that, it knows the grant line has no break.
It probably caused more problems than it caught, which is my guess as to why
no QBUS machine has/uses it.
The -11/34 (not the /34A) has something unusual for grant timeouts, but I
forget the details. I'll look it up.
Noel
