Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!tut.cis.ohio-state.edu!bloom-beacon!usc!orion.cf.uci.edu!uci-ics!zardoz!dhw68k!bob
From: bob@dhw68k.cts.com (Bob Best)
Newsgroups: comp.os.minix
Subject: Re: Evans Minix Kernel - problem report
Summary: problem is with 1.1 compiler w/ fix
Message-ID: <23267@dhw68k.cts.com>
Date: 28 May 89 07:47:28 GMT
References: <23213@dhw68k.cts.com>
Reply-To: bob@dhw68k.cts.com (Bob Best)
Organization: Wolfskill & Dowling residence; Anaheim, CA (USA)
Lines: 35

In article <23213@dhw68k.cts.com> I wrote:
>I have managed to compile the Evans protected mode minix kernel
>using the 1.1 compiler binaries.  I had major problems with /lib/cpp
>on the *.x files which I finally processed using Xenix.
The problem is that 1.1 cpp does not remove C comments from the output.
If you use this binary, you should remove the comments from the .x files
before processing.

>Unfortunately,
>although the kernel boots and runs, it eventually panics and aborts
>with a 'panic - MM can't reply' message.  While the system is running,
>I am receiving messages of the form 'tty_reply return status is ...'
>Bruce, if you or any other hacker out there has a clue, please post
>your suggestions.  I don't trust the code generated by the 1.1 compiler,
>but I have used it to compile all previous versions of Minix.
The problem is in the p2_s_call and s_call functions in mpx286.x and mpx88.x,
respectively.  I will be posting unofficial patches, but the fix requires
saving the si register prior to 'call _sys_call' in both of these procedures.
The 1.1 compiler was NOT preserving the si register on function calls.
For example, use the following:
	push si
	push bx
	push ax
	push cx
	call _sys_call
	add  sp,#6
	pop  si
	mov  AXREG(si),ax

The moral of all of this is quite simple:  Using the 1.1 compiler is
bound to leave you chasing bugs that don't exist.

-- 
Bob Best
uucp: {spsd, zardoz, felix}!dhw68k!bob	InterNet: bob@dhw68k.cts.com