AminetAminet
Search:
81839 packages online
About
Recent
Browse
Search
Upload
Setup
Services

util/boot/AllocP32.lha

Mirror:Random
Showing:m68k-amigaosppc-amigaosppc-morphosi386-arosi386-amithlonppc-warpupppc-powerupgeneric
No screenshot available
Short:AllocP32 - BetterAlloc (AllocMem patch for PPC) V2.01
Author:thor at math.tu-berlin.de
Uploader:thor math tu-berlin de
Type:util/boot
Version:2.01
Architecture:m68k-amigaos
Date:1997-10-16
Download:http://aminet.net/util/boot/AllocP32.lha - View contents
Readme:http://aminet.net/util/boot/AllocP32.readme
Downloads:1546
(with the friendly permission of the original author, Andreas Kleinert)

_____________________________________________________________________________

New in release 2.01:

	- Added a workarount for stupid "SaferPatches" clones.

_____________________________________________________________________________

New in release 2.00:

	- Rewritten in assembly language, much shorter file
	- Tries a partial flush first, instead of removing
	  all libraries on failure.
	- Removed unnecessary AllocVec patch since it calls
	  AllocMem anyways.
	- Added workarounds for bugs in the layers.lib

_____________________________________________________________________________

AllocP32 will be part of a major memory defragmentizer project still to be
published, called PoolMem. Please stay tuned!
_____________________________________________________________________________

WARNING: THIS PATCH ISN'T WITHOUT QUIRKS! READ THIS FILE CAREFULLY BEFORE
INSTALLING ALLOCP32. REALLY, I MEAN IT!
_____________________________________________________________________________

AllocP32 is identical to AllocP EXCEPT that it allocates memory in 32 byte
wide blocks to avoid conflicts between PPC and MC68K microprocessors. If
you don't have a PPC board, this patch is most likely unnecessary and will
only slow down your computer and waste memory. Use AllocP instead.

This program is still in an "experimental" phase, even though it has been
tested now. No guarantee that this works with a PPC board, I don't have
one for testing! (However, it's welcome as a gift.... ;-)
_____________________________________________________________________________

 Sometimes programs fail with a "not enough memory" error,
 but after calling "avail flush" the same operation does
 succeed without problems.

 
 If AllocMem routine in the ROMs did not find enough memory,
 it tries to flush disk based libraries and devices and,
 afterwards, tries again to reallocate the memory.
 However, due to a design flaw of the AllocMem() routine,
 this memory flushing does not have the desired result some-
 times - even though the libraries have been informed to
 remove themselves, the memory is not available directly
 afterwards. The AllocMem() call will fail anyways, EVEN THOUGH 
 the requiested memory will be available immediatly after completion 
 of AllocMem().
 The reasons for this strange behaiviour are rather technical 
 and explained below, for the interested user.


 This patch does ensure that AllocMem/AllocVec won't
 fail unless there's really no memory available, even
 by flushing. It does ALSO, and in addition to AllocP,
 ensure that the allocated memory is contained in 
 32 byte memory blocks and not on boundaries of these
 blocks, so match in one cache line of either the PPC
 or the MC68K microprocessor, not in both. 
 This avoids that parts of the allocated memory is in 
 different caches of the two CPUs and not up-to-date. 
 The patch DOES NOT ensure that the memory is 32 byte 
 aligned! (The current implemenation returns always 
 memory eight bytes on top of a 32 byte boundary, 
 but don't take this for sure!)

 The patch does NOT apply to structures allocated by the
 layer library so take care! The reason is a design flaw
 (avoiding to call this a serious bug) in the layers
 library, which causes also some trouble for MungWall
 and MemSniff. See below for what this bug is about.

WARNING: Due to this workaround of the layers.libary bug,
 it is currently NOT allowed to install an AllocMem()
 patch ON TOP of this one or it will CRASH! If a second
 patch of AllocMem is unavoidable, run AllocP32 LAST!
 To find out which patches are installed in your system,
 get the "SaferPatches" package from my home page at

 http://www.tu-berlin.de/~thor/thor/index.html

 or from the AmiNet.

 Because of the memory alignment, AllocP32 must allocate
 more memory than requested and will "round up" the memory
 size to fit entirely in 32 byte blocks. This DOES, of course,
 waste memory! Don't use this patch without good reason!
 (good reason = a PPC board!)


 Note:  Works now for all operating systems and all CPUs,
	is no longer restricted to V37 or MC'20.
	Does not require a PPC, but if you don't have one,
	try AllocP instead - it's much less memory consuming.

 Usage:  Try starting in the Shell/CLI.
         If it does run stable, copy it into
         your C: directory and add it
         somewhere into your s:user-startup

                AllocP32

 You use this patch at your own risk.
 No guarantee for anything.
 Source code in assembly language included, requires the DevPac assembler
 and my macro package at dev/asm/DvPkMacros.lha in the AmiNet.


 All mentioned trademarks are subject to their owners.
_____________________________________________________________________________

The design flaw in AllocMem():

	When looking closely at the ROM routine of AllocMem(), you'll
	notice that a memory flush is TRIED if the first allocation
	failed. 
	Why does AllocMem then fail, even though if enough memory
	is available?

	Consider the following situation:
	A library has launched a sub-task for control of some of its
	features. This sub-task could be used for disk I-O, for example.

	If this hypothetical library has to be expunged, it can't do
	this on its own because this will release memory of the sub-task
	which is still running. This sub-task must be informed somehow
	that its memory must go. While this is in principle simple to do -
	just send a message to the sub task - the library can't wait for
	an answer of the sub-task because this would brake the Forbid()
	state, which is illegal at this point of the operation.
	The only solution for the library is to leave the memory flush
	to the sub task and to return to the low-memory handler of exec
	WITHOUT trying to remove itself - leaving this to its sub-task.
	
	On the other hand, since multitasking is still forbidden, there
	is NO chance that this subtask will actually get a chance to 
	remove the library, leaving it in memory and causing the 
	allocation failure. EVEN THOUGH the library was informed to
	get removed, there is no chance of doing this while flushing.

	However, as soon as exec leaves the AllocMem routine, multi-
	tasking will be turned on again, thus causing a task switch to
	the subtask and NOW causing the removal of the library - to late
	to have any effect.

	AllocP tries to avoid this situation by calling AllocMem a
	second time.

_____________________________________________________________________________

The bug in the layers.library:

	The layers.library allocates memory for tempory region information
	in big chunks and cuts this memory apart in smaller pieces, freeing
	the snippets individually. This is according to the RKRM standards
	ILLEGAL and has caused a lot of problems (including crashes of
	earlier versions of AllocP32) not only for AllocP32. 
	Please DO NOT share layers.library structures between different CPUs
	until this has found a solution in a bugfixed OS version. If this
	can't be avoided, flush the data caches!

_____________________________________________________________________________

                        The THOR-Software Licence


This License applies to the computer programs known as "AllocP32".
The "Program", below, refers to such program.


The programs and files in this distribution are freely distributable
under the restrictions stated below, but are also Copyright (c)
Thomas Richter.


Distribution of the Program by a commercial organization without written
permission from the author to any third party is prohibited if any payment
is made in connection with such distribution, whether directly
(as in payment for a copy of the Program) or indirectly (as in payment
for some service related to the Program, or payment for some product
or service that includes a copy of the Program "without charge";
these are only examples, and not an exhaustive enumeration of prohibited
activities). However, the following methods of distribution involving
payment shall not in and of themselves be a violation of this restriction:


(i) Posting the Program on a public access information storage and
retrieval service for which a fee is received for retrieving information
(such as an on-line service), provided that the fee is not
content-dependent (i.e., the fee would be the same for retrieving the same
volume of information consisting of random data).



(ii) Distributing the Program on a CD-ROM, provided that the files
containing the Program are reproduced entirely and verbatim on such
CD-ROM, and provided further that all information on such CD-ROM be
redistributable for non-commercial purposes without charge.



Everything in this distribution must be kept together, in original
and unmodified form.




Limitations.

THE PROGRAM IS PROVIDED TO YOU "AS IS," WITHOUT WARRANTY. THERE IS NO
WARRANTY FOR THE PROGRAM, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
PARTICULAR PURPOSE AND NONINFRINGEMENT OF THIRD PARTY RIGHTS. THE ENTIRE
RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD
THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY
SERVICING, REPAIR OR CORRECTION.



IF YOU DO NOT ACCEPT THIS LICENCE, YOU MUST DELETE ALL FILES CONTAINED IN
THIS ARCHIVE.

_____________________________________________________________________________

Thomas,

	October 1997


Contents of util/boot/AllocP32.lha
 PERMSSN    UID  GID    PACKED    SIZE  RATIO     CRC       STAMP          NAME
---------- ----------- ------- ------- ------ ---------- ------------ -------------
[generic]                  455     592  76.9% -lh5- 4b81 Oct 14  1997 AllocP32/AllocP32
[generic]                 3492   12990  26.9% -lh5- f37e Oct 14  1997 AllocP32/AllocP32.asm
[generic]                 4164    9618  43.3% -lh5- c9ab Oct 14  1997 AllocP32/AllocP32.readme
[generic]                  377     856  44.0% -lh5- 76da Oct 11  1997 AllocP32/AllocP32.readme.info
---------- ----------- ------- ------- ------ ---------- ------------ -------------
 Total         4 files    8488   24056  35.3%            Oct 16  1997

Aminet © 1992-2017 Urban Müller and the Aminet team. Aminet contact address: <aminetaminet net>