Check checksum on CH11 under KLH10 #2342
Conversation
|
This triggers a HALT instruction on machines other than KS10. EDIT: SIMH KS10 also fails, although V3 works. Strange. Anyway, the missing |
src/system/chaos.291
Outdated
| @@ -2426,29 +2426,49 @@ BBLK | |||
| JRST [ AOS CHNLOS ;No buffers, lose this packet | |||
| JRST CHXBK1 ] | |||
| HRLI A,-%CPKDT ;Copy out the packet header | |||
| IFN KLH10P,[ | |||
There was a problem hiding this comment.
This needs a closing ]. Or the line could just be removed because the comment doesn't need to be inside a conditional.
src/system/chaos.291
Outdated
| ;; this code should be taken out again. | ||
| ;; Note that the performance impact is very small. | ||
| ;; /BV November 2024. | ||
| ;; (Side note: is "ADDI J,(B)" faster than "ADD J,B"?) |
There was a problem hiding this comment.
On some machines, potentially yes. The reason being that the immediate operand can use the possibly streamlined 18-bit addressing circuitry. I believe it is a little faster on a KA10, or maybe a PDP-6. Certainly this type of optimization can be seen in early MIT code.
There was a problem hiding this comment.
I don't think ITS source code comments are the right place for a Q and A, though.
larsbrinkhoff
force-pushed
the
bv/klh10-ch11-checksum-check
branch
from
4d3613f
to
0989784
Compare
I wish MIDAS would complain about unbalanced |
It makes some half-hearted attempts. But in this case, I think it did complain: That's the line which .INSRTs NET, which in turn .INSRTs CHAOS. So hardly a great clue, but still, something. |
There is some bug in the CH11 implementation for KLH10 which makes it corrupt packets - quite rarely, but making e.g. remote DUMPs (using the RTAPE protocol) unreliable. I've previously established that it must be the CH11, rather than cbridge or the RTAPE server, that is to blame. My hunch is that there is some synchronization problem in the CH11 implementation.
The CH11 implementation in KLH10 is based on the IMP implementation, and I did it some 20-25 years ago. It uses really hairy mechanisms for synchronization between the KLH10 end (dvch11) and the device process (dpchaos). Hacking this type of C code is far from my favourite thing to do, so I've been putting it off since there are so many other things to hack that are much more fun and rewarding.
This PR provides a workaround: when running on a KLH10 that uses CH11, ITS computes the checksum when a packet has been received, and thus discovers corrupt packets and simply drops them (after counting it which will appear in STATUS replies). The performance price is really low.
The underlying bug in KLH10s CH11 should be fixed at some point, but that's a separate issue (and not for this repo).
Side note: is "ADDI J,(B)" (significantly) faster than "ADD J,B" on a KLH10?