Path: csiph.com!weretis.net!feeder8.news.weretis.net!eternal-september.org!feeder3.eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Colin Leroy-Mira <colin@colino.net>
Newsgroups: comp.sys.apple2.programmer
Subject: Help me with a crash theory?
Date: Tue, 23 Apr 2024 09:04:59 +0200
Organization: A noiseless patient Spider
Lines: 57
Message-ID: <20240423090459.6e9a34b4@laptop-sigfox>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Injection-Date: Tue, 23 Apr 2024 09:05:02 +0200 (CEST)
Injection-Info: dont-email.me; posting-host="6f48bc08933f1ed7594ac86d2b3c7c06"; logging-data="1575461"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1/n0dHb2yJPgxumFhm8ZE43FCKsiHfElUw="
Cancel-Lock: sha1:vvAn23EKTGjlrly8nz4Njlnhm1k=
X-Newsreader: Claws Mail 3.19.1git30 (GTK+ 2.24.33; x86_64-pc-linux-gnu)
Xref: csiph.com comp.sys.apple2.programmer:6309

Hi all,

I have a low-level ACIA serial question for you. Context: I'm working
on my Wozamp thingy, specificaly the audio+video streaming part. It
works similar to A2stream for the audio, except it reads from the ACIA
instead of an Uthernet card to redirect execution to the next duty
cycle.

I have started investigating a new, weird bug yesterday evening. I had
relocated my duty cycle code, which consists of 32 different duty cycles
functions, all aligned on pages so I can jump from one to another by
updating just the high byte of the jmp pointer.

My duty cycle functions started at $6000, $6100, etc to $7F00 and
everything worked fine. Yesterday I relocated that to $6400 to
$8300. Emulation still works good, but it crashes on real hardware.

The only theory I have right now is that it's due to my savage way of
reading the next destination byte from the ACIA DATA register: I don't
have enough cycles to check the STATUS register, so I just read the
DATA one. At worst, I thought, I'll re-read the same byte, and that's
not a problem. (and it was not a problem with my functions from $6000
to $7F00). But in fact, I'm wondering, can I read a byte from the DATA
register in the middle of the ACIA writing it? This would explain the
bug:

$60 = 01100000
$7F = 01111111
But:
$64 = 01100100
$83 = 10000011

If I indeed read a byte right when it's written, the first solution is
no problem, because the high bit is always 0, and then the ACIA can
update bits in the order it wants, I have a handler at any 011xxxxx
location.

But the second solution would be problematic: what if the ACIA updates
the data register from $7E (01111110), to $81 (10000001) ? If the high
bit is the first one updated, I could read 11111110 ($FE), and jump to
$FE00 which would be bad.

Or, if I go from $83 (10000011) to $64 (01100100) ? Reading in the
middle of this write could result in 00000011 ($03, bad) if bits are
written high to low, or 1000100 ($84, also bad) if they are written low
to high.

In both cases, reading the DATA register without a care for the
STATUS register telling me a byte is ready just happens to work safely
with numbers from $60 to $7F, but can wreak havoc with numbers from $64
to $83.

Does this theory sound plausible to you folks?
-- 
Colin
https://www.colino.net/