New in FreeMiNT 1.15.0 (short list):

- integrated FAT/VFAT/FAT32 filesystem
  - read TOS & DOS partitions of any size;
    support FAT32 partitions larger than 2 GB
  - very fast
  - support the VFAT extension for easy data exchange
  - mostly configurable at runtime
- integrated block cache
  - efficient memory optimising LRU algorithm
  - supported from MinixFS 0.70 and NEWFATFS
  - features write back mode with drive granularity
    and runtime configuration
- high resolution timer
  - 26 microseconds granularity
  - timezone support
- improved Ssystem (see Ssystem documentation)
  (timezone support, sync time and a lot of other things)
- improved SecureLevel support
  (XBIOS is now protected from user calls)
- Milan patches (extra kernel)
- ESCC patches (extra kernel)
- complete 060 patches
- internal speed optimizations
- many minor patches

- new configuration keywords for MiNT.CNF:

NEWFATFS=<comma separated drive list>
# enable NEWFATFS for specified drives

VFAT=<comma separated drive list>
# enable VFAT extension
# used by NEWFATFS

WB_ENABLE=<comma separated drive list>
# enable write back cache for specified drives
# used by NEWFATFS and MinixFS 0.70

CACHE=<size in kb>
# increase filesystem cache
# used by NEWFATFS and MinixFS 0.70

HIDE_B=
# remove drive B: from drive list

- completly resorted source
- global headerfiles
- a lot of new header files for better programmer support

Take a look on Changes.MH for a detailed technical description
of all changes.

Refer to ssystem.doc, block_IO.doc, fatfs.doc, the included examples
and the source distribution for details and exact descriptions.

Changes were contributed by the following people (alphabetical order):
  - Thomas Binder <gryf@hrzpub.tu-darmstadt.de>
  - Katherine L. Ellis <kellis@primenet.com>
  - Christian Felsch <felsch@tu-harburg.de>
  - Guido Flohr <gufl0000@stud.uni-sb.de>
  - Konrad M. Kokoszkiewicz <draco@mi.com.pl>
  - Mikko Larjava <mitala@nic.funet.fi>, 68060 cache patch (once more)
  - Daniel Petersson <tam@dataphone.se>
  - Julian F. Reschke <fjr010@email.mot.com>
  - Rainer Seitel <Rainer_Seitel@ka.maus.de>

Frank Naumann <fnaumann@cs.uni-magdeburg.de>
Magdeburg, den 8.10.1998

======================================================================
======================================================================
======================================================================

Differences between 1.14.6 and 1.14.7:

- pipefs.c fix for bidirectional IPC.
- now it is possible to chdir to a directory that is set as unreadable
  (--x--x--x).
- Videl patch for sync errors on Falcons has been applied. Now vcons1d doesn't
  produce broken display after switching to a virtual console.
- TOS 4 bug referring to the FastRAM has been worked around. NOTTRAM.PRG is no
  longer needed in the AUTO folder.
- cache handling has been patched for 68060. Now the 060 is considered a '040-alike',
  not '030-alike', as it was before.
- single mode (no multitasking) has been added. The scheduler can be stopped so that
  the calling process is not preempted until it unlocks the scheduler back to normal
  operation.
- "live" timeslice control has been added. You can change the number of slices without
  a reboot (its actually the same value as that set by the SLICES keyword in the MINT.CNF).

Refer to changes.mh, ssystem.doc, included examples and the source distribution for
details.

Several more TOS calls have been restricted to euid root at security level 1, as follows:

Call		Reaction on a call from an unprivileged process
----		-----------------------------------------------
Midiws()	EACCDN
Iorec()		termination
Rsconf()	termination
Bconmap()	EINVFN
Dosound()	EINVFN
Vsetmode()	EINVFN
Vsetscreen()	EINVFN
Setscreen()	EINVFN

Generally, protected calls return an error code wherever it is possible. If the function
operates such a way, that calling programs don't check the returned value for an error
condition, or the function is intended to return a memory pointer the further program
execution may rely on, the caller is terminated.

Changes were contributed by the following people (alphabetical order):

- Bernd Ebert
- Katherine Ellis
- Guido Flohr
- Maurits van de Kamp
- Konrad Kokoszkiewicz (me)
- Torsten Lang
- Kristoffer Lawson
- Martin-Eric Racine

I apologize for lack of patches referring to more general problems, like still
broken virtual memory manager or memory fragmentation.

No warranty, use at your own risk.
Draco

Warszawa, March 1998.

----------------------------------------------------------------------
Notes for FreeMiNT 1.14.6 beta:

Differences between 1.14.6 and the previous 1.14.5:

- security levels introduced (actually 0, 1 and 2)
- kernel functions Setexc(), Rwabs() and Dwritelabel() are restricted 
  for euid root at level 1.
- kernel functions Supexec() and Super() are restricted for euid root 
  at level 2.
- the keyword "SECURITY=YES|NO" in MINT.CNF has been replaced with
  more flexible SECURELEVEL=n (where n is a value of 0, 1 or 2).
- new system call, Ssystem() has been introduced to the kernel.

Kernel protection is still unfinished, particularly no XBIOS function 
is protected except Supexec(). Functions restricted for euid root, 
when called by an unprivileged user while the appropriate security 
level is set, behave two different ways:

- calls like Rwabs() and Dwritelabel() will return status EACCDN 
  ("Access denied").
- calls like Setexc(), Supexec() and Super(), those don't use any 
  status code normally (i.e. there's no way to signalize an error),
  will terminate (kill) the calling process.

Actually you can safely use security level 1 in any configuration. 
Level 2 however (which restricts Super() and Supexec() for root 
processes) can't be currently used on multiuser setups because of the 
way the MiNT Library (which is actually linked to 99.9% Unix programs 
ported to MiNT) works. That means that to take advantage of full 
kernel protection, the MiNTLib must be fixed and the whole Unix stuff 
has to be recompiled. That's a lot of work and it will take some time.

However, the kernel now makes it possible. MiNTLibs use supervisor 
mode to access Cookie Jar and some GEMDOS variables located in the 
first 32k of address space. The kernel now supports such operations, 
so there's no more need for user processes to switch to supervisor 
mode. That was the reason to introduce the Ssystem() call. Its 
purposes:

- read GEMDOS variables from the first 32k of memory
- set GEMDOS variables (root only)
- get existing Cookie Jar entries
- place new Cookie Jar entries (root only)
- get some kernel information

See system2.c from the source distribution and included examples for 
details and use.

The original code for Ssystem() has been developed by Jerry Geiger. 
The idea has been redesigned by me. No warranty as ever, use at your 
own risk.

Enjoy,
Draco
draco@mi.com.pl

Warszawa, 7.XII.1997.

----------------------------------------------------------------------

Notes for FreeMiNT 1.14.5 beta:

Since FreeMiNT 1.14 has been released, it turned over, that Profesore's VM
extension doesn't work stable on many TT computers and doesn't work at all
on Falcons. From that time, VM extension has got some dedicated patches
and fixes and it started really working. However, before the VM has been
gotten to work, I decided to split the 030 kernel into two versions. This
left until now, as somebody may prefer non-VM kernel (which is actually
more stable than the other with VM enabled). There are three supplied
binaries:

mint.prg - 68000 version
mint030.prg - 68030 version
mintvmx.prg - 68030 version with VM extension

That last may be used for some experiments which (I hope) would finally
lead to fix the VM completely. If you have suggestions, please contact me
directly (draco@mi.com.pl) or via MiNT mailing list.

Notice: even if you use the VM kernel version, but name it MINTNM.PRG in
your AUTO folder, the VM extension will be disabled.

OK, here are main differences between 1.14.5 and original 1.14:

- VM is fixed so that it can (hopefully) initialize & start on any
  030-based computer with any drive and any hard disk driver. Tested on a
  Falcon030, 14 MB ST RAM, HDD Conner CFS 1621A (IDE) and HDDRIVER
  (NOTICE: it seems to work *much* better than before, but still isn't
  100% stable)
- PMMU cookie will be set in the Cookie Jar when VM is active.
- the Cookie Jar has been increased and has now 8 free entries more.
- serial patch is applied so that data transfers up to 230000 baud via 
  RS-232 seem to be possible.
- pexec() patch is applied to avoid possible crashes on 040-based machines
  while loading filesystems (XFS).
- unaligned block memory moves (quickmov.spp) should be now faster on 030.
- new command, SECURITY, is introduced to the MINT.CNF. Default is
  SECURITY=NO. If set to YES, Rwabs() and Dwritelabel() are available
  for superuser (root) only.
- a bug in procfs.c is fixed (function proc_remove()), so that regular
  users are no longer able to remove processes those aren't their own
  (it was known as "ftpd /proc bug").

Security oriented stuff has been experimentally introduced as a result of
long and sometimes hot discuss on the MiNT mailing list. It is intended to
prevent regular users, who don't have root privileges in multiuser
configurations, from doing ugly things to the system. For now, only
Rwabs() has been protected so that any user but root has no possibility to
reformat the drive. Also Dwritelabel() will be denied for user processes. 
Please test these new features carefully and report any found problems to
either MiNT mailing list (mint@atari.archive.umich.edu) or directly to me
(draco@mi.com.pl). 

Note it is planned for the future to fix Super() and all XBIOS functions
so that they would be superuser-only, when the SECURITY option is enabled.
However, it needs additional kernel functions, those are currently under
development. Super() and XBIOS will work traditional way until new
MiNTLibs and the complete new MiNT distribution would be prepared.

If you run MultiTOS or any other non-multiuser setup, please set
SECURITY=NO.

Serial patch has been done by Torsten Lang (Torsten.Lang@jung.de). Pexec
patch has been done by Harry Eaton (haceaton@aplcomm.jhuapl.edu). The
responsibility for all other changes (see changes.mh) and this kernel
version is taken by me (draco@mi.com.pl). No warranty as ever, use it at
your own risk.

Enjoy,

Draco,
Warszawa, 13.IX.1997.

PS. Binaries are compiled with gcc 2.7.2

****************************************************************************

This is FreeMiNT 1.14 BETA:

FreeMiNT 1.14 has virtual memory which use a dedicated 
swappartition. The swappartition is created by the doswap
utility included in the distribution.

MiNT now also have a new recognised "nameparameter". If you rename
mint.prg into mntnm.prg or mintnm.prg and run it from the autofolder
then both memory protection and virtual memory will be disabled.
Virtual memory is only enabled if memory protection is disabled,
you have a swappartion installed,you have enough memory in your 
computer and have a true 68030 in your computer.
Memory protection is turned off by for example renaming mint.prg
into mintnp.prg and running it from the autofolder.

VM will not work with 040,060 or 020 with 68851 coprocessor. It
will only work with a true 030 with mmu. It will not work with a
68EC030 or similar processor.

WARNING: The specific type of 030 is not tested. The VM will try
to run and will crash on a EC030. If you have such a processor
you should always run MiNT with both memory protection and VM
turned off!

I strongly suggest you to make a backup of your harddisk before
experimenting with VM. Bevare that this version is a beta and is
not tested on all systems. It is tested on a plain 8 MHz 2.5 meg
520ST and a TT clocked at 32 MHz with 4 meg ST-ram and 4 meg TT-ram
using a Quantum 850 meg Trailblazer scsi harddisc fitted to the internal
TT scsibus and using AHDI V5.00 as harddisc driver. It is not tested on
any other machine and may perhaps not run on other machines. Although
that is highly unlikely.

VM larger than 64 meg is not supported and doswap will complain
if you try to install a larger swappartition than 64 meg.

By default all TT-ram and all ST-ram but 4 meg is used as page
memory ie only 4 meg ST-ram will be left. This can be changed
if the symbol ST_LEFT_AFTER_VM in mem.h are changed to the amount
of ST-ram needed. On a TT you can set this as low as 1 or 2 megs if
the program flags on your applications are properly set. On a Falcon
you might want to have more ST-ram free.


Version 1.14 (ie VM extension) was made by Sven 'il Profesore' Karlsson.
Patches from 1.12.5 to MiNT 1.12.6 were collected by Charles 
Briscoe-Smith <cpb4@ukc.ac.uk>.  Versions of FreeMiNT up to and 
including 1.12.5 were collected by Michael Hohmuth 
<hohmuth@inf.tu-dresden.de>

21th of September 1996


/*
 * This file belongs to FreeMiNT.  It's not in the original MiNT 1.12
 * distribution.
 */

This is an attempt in helping to re-synchronize the MiNT source tree
between the MiNT Mailing List folks.  This effort has previously been
known as "MH-MiNT".

Please note that FreeMiNT is unsupported.  There is no warranty.  See
the file COPYING for details.

Please report bugs to the MiNT mailing list <mint@atari.archive.umich.edu>

The supplied binary is compiled with gcc 2.3.3.

WARNING: This is only a beta. There is a binary version in this distribution
but it is only meant for people who are willing to risc their harddisc in the
quest to find all bugs. The VM extension took 3 days to write. It has been
tested only on the machines listed above during 2 weeks. There are probably
configurations that will lead to problems. You should backup all your vital
data before trying this piece of software. It is stable on my TT but I do not
take any responsibility for any lost data or damaged media or hardware.

NOTE: The reltime clock in my TT has stopped working. Since I have not fixed
it yet a lot of files have strange timestamps.

Regards Sven Karlsson, Istari Software
