Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!rutgers!cbmvax!valentin
From: valentin@cbmvax.commodore.com (Valentin Pepelea)
Newsgroups: comp.sys.amiga.tech
Subject: Re: resource tracking
Message-ID: <9716@cbmvax.commodore.com>
Date: 19 Feb 90 02:35:51 GMT
References: <355.25C92297@weyr.FIDONET.ORG> <131188@sun.Eng.Sun.COM>
Reply-To: valentin@cbmvax.cbm.commodore.com (Valentin Pepelea)
Organization: Commodore, West Chester, PA
Lines: 53

In article <131188@sun.Eng.Sun.COM> cmcmanis@sun.UUCP (Chuck McManis) writes:
> 
>Something I personally would like to see would be a "runprotected" command
>that would start a program and setfunction the appropriate vectors so that
>the program would run with the MMU protecting all addresses outside of it's 
>range.
>
>It need only protect them from Write Access and could stop the 
>program when it did something stupid. Too bad Valentin already has
>his degree this might be a good thesis project :-) 

Not that I did not think about it. The problem is that I can not achieve full
recoverability if a task goes astray. The problem lies with the definition
and useage of MEMF_PUBLIC. Memory defined as such is free to be written to
by any task or interrupt in the system. That's where all the system structures
are.

So if one task crashes, even if the MMU protects other private memory, it could
be that a portion of MEMF_PUBLIC memory got corrupted by some other task, and
that eventually resulted in the crash under the context of the current task.

Full recoverability in a memory protected system can therefore be achieved only
by changing the definition of the MEMF_PUBLIC flag, and by adding parameter
checking to all system functions that cause another task or interrupt to
execute some code. For example the DoIO() function causes a device to process
a request. If the request provided is scrambled, a crash will occur under the
device's context.

Both these changes would cause some programs to fail, if implemented. So for
backwards compatibility reasons, it is unlikely that memory protection will
ever be implemented on AmigaDOS. I for one would gladely implement them as they
require no great engineering effort, but the god in the corner office is 
unlikely to approve, for backward compatibility reasons.

>This would be a great debugging tool and allow a developer to 
>release it with confidence on the unsuspecting user community
>*knowing* that it doesn't do weirdo writes outside of its address
>space.

As a debugging tool, "runprotected" would still make sense, since many illegal
memory accesses would be detected this way. In fact, we already have two such
utilities in-house. Both trap accesses to page zero (0x000000-0x000100) and
above 16 Meg, send out a status report on the serial port at 9600 baud, and
then complete the memory accesses manually. One was written by Bryce and the
other by me. We like so much to double engineering efforts, you see. Of course,
mine is muchg better written.  :-)

Valentin
-- 
The Goddess of democracy? "The tyrants     Name:    Valentin Pepelea
may distroy a statue,  but they cannot     Phone:   (215) 431-9327
kill a god."                               UseNet:  cbmvax!valentin@uunet.uu.net
             - Ancient Chinese Proverb     Claimer: I not Commodore spokesman be