27 Feb
2002
27 Feb
'02
6:16 a.m.
On Feb 27, 12:21, Andreas Freiherr wrote:
Darn! - I disassembled the whole stuff manually
yesterday evening!
:-) I apologise for the slightly non-standard syntax used, notably for
octal constants; it was rewritten for (and then assembled by) a home-grown
cross-assembler.
Non-obvious is that only two sectors (1 and 3) need to
be read: the code
in the manual continues up to sector 7 (constant in word 2124, or in
your copy, at 1124), in steps of 2 (BTW, why are even sectors skipped?).
I think there were actually three errors in my original copy. Anyway, the
version on my website is now correct. It's not the only way to do it, I've
seen other versions (there are similar versions, not identical, but
interestingly enough with mostly the same errors, in various RT-11 manuals,
the Microcomputer Peripherals Handbook, and the RX02 Pocket Service Guide).
The reason for the number of sectors is that you can use the same bootstrap
with trivial alterations for a single-density boot, but then you need four.
sectors, not two. The reason for missing the even-numbered sectors is
because they are software interleaved to give time for the memory transfers
between reads.
The most non-obvious error is using the word at
location 1132 either as
a HALT instruction (when aborting bootstrap due to an error detected by
one of the BMIs) or as a constant to load into R0 (when finishing
bootstrap and passing control to address zero, for which a CLR PC is an
excellent JMP @#0 replacement). My copy of the LSI-11 manual had this
completely wrong - it would fail to abort properly.
Some versions actually have the HALT elsewhere. I once spent ages trying
to work out the shortest variation on this bootstrap, IIRC I managed to
save just one word.
Starting the bootstrap might be easiest with a 1000G
(or in the original
copy, 2000G). one command shorter than setting R7 to 1000 and then
<P>roceeding.
The reason for the <P> instead of <G> is to avoid a bus reset, which makes
the RX02 and some processors "do things". And you're supposed to disable
interrupts by setting a mask in RS first. It doesn't use the stack,
though, and any decent second-stage boot should set R6, so I suspect that's
redundant.
I'm surprised your version starts at 2000. By convention, all normal
bootstraps and similar start at 1000 -- the best-known exceptions being
XXDP code (starts at 200, same as on a PDP-8) and the TS11 bootstrap which
starts at 10000 (to allow space for larger tape blocks).
--
Pete Peter Turnbull
Network Manager
University of York
27 Feb
27 Feb
7:48 a.m.
New subject: RX02 Bootstrap
Pete,
:-) I apologise for the slightly non-standard syntax
used, notably for
octal constants; it was rewritten for (and then assembled by) a home-grown
cross-assembler.
No problem: since we are interested in the sequence of words in octal,
it doesn't matter too much how you format the right side of your
listing. It's well readable and self-explanatory anyway.
I think there were actually three errors in my
original copy. Anyway, the
Yes, the place I am referring to also has three: one is at 2066 (1066 in
your version), where the value is erroneously printed as "1100420" and
should be "100240" (the difference in the low-order bit has to do with
error #3).
The second error is at 2114, where it shows "6052" and should be
"60502"
(this is one of the two ADD R5,R2 instructions to advance to the next
memory address).
And the third is that the branch offsets in all BMIs yield a destination
address of 2130, where instead of aborting the boot process it would
activate the routine that has not yet been read completely. Ouch!
Also, the last three locations are slightly different, but this
shouldn't present a problem:
2130 005000 CLR R0
2132 005007 CLR PC (my favourite PDP-11 instruction, really!)
2134 000000 HALT
All this is taken from "LSI-11 Systems Service Manual, Volume III", DEC
order no. EK-LSIFS-SV-005. These digits are "Fifth Edition, January
1985".
The reason for the number of sectors is that you can
use the same bootstrap
with trivial alterations for a single-density boot, but then you need four.
Aaah! - Good reason.
sectors, not two. The reason for missing the
even-numbered sectors is
because they are software interleaved to give time for the memory transfers
between reads.
I thought this is done by spacing the sectors on the disk surface (like
1, 13, 2, 14, etc.) and having drive electronics deal with the sector
number from the preamble? Consequently, this would be a property of the
floppy that would be implemented when the media was formatted.
Some versions actually have the HALT elsewhere. I
once spent ages trying
to work out the shortest variation on this bootstrap, IIRC I managed to
save just one word.
I think more significant savings could perhaps be obtained by
initializing a stack at the beginning and then using a subroutine like
WAITRX: BIT R0, (R1)
BEQ $-4
BMI ABORT
RTS PC
ABORT: HALT
to wait for any of the TR, ERR or DONE bits to come up, instead of
repeating these instructions every time. You might even free up R0 and
use the constant #100240 directly in the BIT instruction, because you
won't need it elsewhere any more.
The reason for the <P> instead of <G> is
to avoid a bus reset, which makes
the RX02 and some processors "do things". And you're supposed to disable
interrupts by setting a mask in RS first. It doesn't use the stack,
though, and any decent second-stage boot should set R6, so I suspect that's
redundant.
Since you never set the INTR ENB bit in the RX2CS register, you just
need to keep your fingers crossed so no other device will issue an
interrupt request... ;-)
For a ROM-based implementation, I'd certainly agree that it is wise to
make sure no interrupt will get in the way, and it's certainly safer to
initialize the stack pointer as soon as possible.
I'm surprised your version starts at 2000. By
convention, all normal
bootstraps and similar start at 1000 -- the best-known exceptions being
XXDP code (starts at 200, same as on a PDP-8) and the TS11 bootstrap which
starts at 10000 (to allow space for larger tape blocks).
Yes. But this is neither a typo nor a printing error. If you read four
blocks of 200(8) _words_ each, starting at zero, you fill exactly
2000(8) _bytes_, and the next free location is 2000(8). As you state,
the printed version is prepared for conversion to single density (by
clearing the 400 bit in locations 2036 and 2072, right?) by reading in
four blocks... - So, the reason is somewhat similar to that for the
TS-11.
--
Andreas Freiherr
Vishay Semiconductor GmbH, Heilbronn, Germany
http://www.vishay.com
8334
days inactive
8334
days old
1 comments
2 participants
participants (2)
-
Andreas.Freiherr＠Vishay.com -
pete＠dunnington.u-net.com