Subject: Writebuffer flushing wrong on 3MIN, MAXINE?
To: None <>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: port-pmax
Date: 12/13/1994 00:18:07
Dean Huxley ( writes:
>Currently, the "cat bootblocks miniroot > /dev/rz#c" method works for
>almost everyone, regardless if they have Ultrix or not.
>Unfortunately, it seems that the bootblocks may not work on certain models.
>        Can't boot (apparently):
>        5000/25
>        5000/120
>        Can boot:
>        2100,
>        3100
>        5000/200 jonathan@DSG.Stanford.EDU
>        (there are other people, but I don't know what they are running)
>If there is anyone that IS running NetBSD native on a 5000/25 or 5000/120
>please speak up.  Also, if anyone else wants to be added/removed from the
>above lists, send me e-mail.

You can add me to the 3100 list too.

Duh.  I know from work on another OS that the 5k/240 aka 3MAX-PLUS has
an I/O asic, with a magic address that needs to be read to flush the
write buffer.  And the 3min (5k/1xx) and Maxine (5k/xx) also have a
similar, but different, I/O asic.  I wonder if the extant
writebuffer-flush works on those machines?  My instinct is, it doesn't

If it doesn't, then probing the TurboChannel might not work as
expected.  And, from work on the 5k/240, resetting the memory errors
after sizing memory certainly doesn't work unless the write buffer is
flushed. And at least on _some_ DECstations, reading from nonexistent
slot addresses causes a memory error that needs to be cleared.

My guess is that MachEmptyWriteBuffer() needs to do something specific
for different models of DECstation, like reading from or writing to a
specific physical addres.  I don't have hw docs for other than 3100s,
5k/200s, and 5k/240s, so I can't say for sure.

Maybe the Sprite source is a good place to double-check this?

I'm busy with a paper deadline and grading till late January.
But if all else fails I can see if I can borrow or steal a 5k/25 for an
hour or so over Xmas and try out one of the bootable disks I already
have.  But it'd be great if someone could disprove this hypothesis