Date:	Fri, 7 Jan 94 23:52:27 PST
From:	stevev AT miser DOT uoregon DOT edu (Steve VanDevender)
To:	DJ Delorie <dj AT ctron DOT com>
Cc:	astor AT alkymi DOT unit DOT no, djgpp AT sun DOT soe DOT clarkson DOT edu
Subject:	Re: What is the address of the code?

DJ Delorie writes:

 > WARNING: Because of the way the cache works on the 386 and 486,
 > opcodes you modify may not be reflected in the cache and won't do what
 > you expect.  The "jmp $+2" trick does NOT work with the 486.  See
 > libsrc/c/dos/int21x.s for an example of self-modiying code that does
 > work.

The 386 has no internal cache, but has a prefetch queue.  The 486
internal cache should not cause any problems for self-modifying
code all by itself; since instructions and data are cached the
same way, modified code in a cache line will be executed as
expected.  It is my understanding that it is the instruction
prefetch queue on the 8086, 80286, and 80386 that makes writing
self-modifying code more difficult, because the prefetch queue
does not act like a cache and modifying instructions that have
already been fetched will not have any effect.  Offhand I do not
know if the 486 has a prefetch queue in addition to its internal
cache; it seems that the cache would make a prefetch queue
unnecessary, though.

- Raw text -