20190211

Strange Code

Now That's Weird

I was trying to compile some ancient code I pulled off the net. It is related to the Venix stuff I've been doing on and off of late.
put = bp->b_nleft;
if (put > cnt)
    put = cnt;
bp->b_nleft -= put;
to = bp->b_ptr;
asm("movc3 r8,(r11),(r7)");
bp->b_ptr += put;
p += put;
cnt -= put;
goto top;
So that's weird, right. What the heck is that movc3 doing in the middle of that code.

This code originally ran on BSD 4.1. The only system that version of Unix ran on was a VAX (later versions were more widely ported, but 4.1 was more of a limited distribution version). OK, looking up the movc3 instruction on vax references online, we see it is the "Move Character" instruction. r8 is the length. srcaddr is (r11) and dstaddr is (r7). So in effect, someone has done an inline of bcopy() here. Now, that's half of the problem. The other half is puzzling out what is in r7, r8 and r11 at the time of this call. In a perfect world, I'd just crank up the compiler to tell me. We live in an imperfect world where spinning up a 4.1 BSD system takes a substantial amount of time.

Fortunately, we can guess. cnt -= put gives us our first clue. We're decrementing by how much we copied, it seems. So r8 (the length) is put. OK. Now, we have this nice variable named 'to' that was most likely in the dstaddr (so r7), and we update it after (to appears only to be here for the side effect, so that's nice). But what's the from? The only thing it could logically be is 'p' since we += it by put as well.

So my best guess is that can be replaced by memcpy(to, p, put); and life will be good. My spidy sense also tells me that we don't need memmove here because they aren't overlapping ranges.

No comments: