This appendix contains instructions to import and install RVM. The process is set up to be a compatible as possible with both Mach 2.6 and Unix BSD 4.3. Support for other versions of Unix is limited, but in general, RVM attempts to keep its system requirements to a minimum to enhance portability. Where necessary for specific versions of Unix and related systems, special instructions are included below.
Please note that RVM is copyrighted by Carnegie Mellon University, and the terms of distribution are specified by the following copyright notice:
RVM: an Experimental Recoverable Virtual Memory Package Release 1.3 Copyright (c) 1990-1994 Carnegie Mellon University All Rights Reserved. Permission to use, copy, modify and distribute this software and its documentation is hereby granted (including for commercial or for-profit use), provided that both the copyright notice and this permission notice appear in all copies of the software, derivative works or modified versions, and any portions thereof, and that both notices appear in supporting documentation, and that credit is given to Carnegie Mellon University in all publications reporting on direct or indirect use of this code or its derivatives. RVM IS AN EXPERIMENTAL SOFTWARE PACKAGE AND IS KNOWN TO HAVE BUGS, SOME OF WHICH MAY HAVE SERIOUS CONSEQUENCES. CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR ANY DAMAGES WHATSOEVER RESULTING DIRECTLY OR INDIRECTLY FROM THE USE OF THIS SOFTWARE OR OF ANY DERIVATIVE WORK. Carnegie Mellon encourages (but does not require) users of this software to return any improvements or extensions that they make, and to grant Carnegie Mellon the rights to redistribute these changes without encumbrance. Such improvements and extensions should be returned to Software.Distribution@cs.cmu.edu.
This release corrects some error reporting bugs in
rvmutl
and the segment loader. No reinitialization of
log files is necessary. All that is required to use this version is
to recompile and relink your applications after importing and
installing RVM.
This release corrects a few small bugs and synchronizes the export and Coda versions. Re-initialization of your log files is recommended.
To reinitialize log files, for each existing log file (or
partition), use your present version of
rvmutl
to
force any remaining records to be reflected in the modified
segments, and then reinitialize the log with the new version of
rvmutl
:
.../RVM1.2/bin/rvmutl # RVM 1.2 * open_log < log file name > * recover * quit
.../RVM1.3/bin/rvmutl # RVM 1.3 * init_log < log file name > < size > * quit
This release contains a more complete verion of the manual and
corrects a bug in
rvm_basher
that caused reports of
virtual memory not matching the disk. The problem was in the test
program itself, not the RVM library.
Other changes are addition of error checking in the segment
loader and RDS initialization functions. Also, the utility
rdsinit
has been made somewhat more forgiving. You can
now enter lengths in decimal as well as octal and hexadecimal. The
values input are checked for minimal agreement with the
requirements of RDS and a size check on the file being created is
made to insure that the regions declared will fit in the file.
No recompilation or log re-initialization is necessary if you are upgrading from the 14 March 1994 release. Those who are using the pre-public release, RVM 1.2, should follow the instructions below.
This release was the first public release, but for those presently using the informal release, RVM 1.2, recompilation of modules using RVM and re-initialization of the log is necessary. To reinitialize the log, use the procedure specified above.
To import RVM, use FTP to get the file RVM.TAR into the directory from which you wish to build RVM. Use the CMU Computer Science FTP server:
cd < your directory > ftp FTP.CS.CMU.EDU Name: anonymous password: < userid > @host cd project/rvm-releases get README # this text get rvm_manual.ps # RVM manual (PostScript) binary # required for tar files get RVM1.3.TAR # packed source files quit
You should also send mail to rvm@cs.cmu.edu to be placed on a mailing list for future updates.
Then unpack the files:
tar xf RVM1.3.TAR
At this point, you can delete the RVM1.3.TAR file to recover disc space.
If you are building for any of the following machine and/or types, you must make the following modifications before continuing the installation. If your machine/OS is not mentioned, no special action is needed.
Delete the following files:
.../RVM1.3/Makeconf .../RVM1.3/rvm/Makeconf .../RVM1.3/rds/Makeconf .../RVM1.3/seg/Makeconf .../RVM1.3/tests/Makeconf .../RVM1.3/bench-SOSP93/Makeconf
make doesnt support the include directive, so you will have to modify the Makefiles. In the following files, replace the include command after the "get basic definitions" with the definitions and macros in the build_defs file which will be found in the RVM directory created by unpacking the tar file:
.../RVM1.3/rvm/Makefile .../RVM1.3/rds/Makefile .../RVM1.3/seg/Makefile .../RVM1.3/tests/Makefile .../RVM1.3/bench-SOSP93/Makefile
Execute the following commands:
cd RVM1.3 make initialize
The above sequences will create the necessary directory
structure rooted as
RVM1.3
in the directory where the
tar file was unpacked, and compile the system. When complete, you
will have the following directories under RVM1.3:
repository for the RVM, RDS and SEG libraries.
repository for RVM-related programs, rvmutl, rdsinit, and the test programs.
repository for header files for RVM subsystems.
source code for RVM.
source code for RDS, a dynamic allocator.
source code for SEG, a loader for RVM segments.
source code and test files for RVM.
source code for the benchmark used in the SOSP-93 paper ].
(Mach only) directory structure for machine-specific build areas, normally cleared after the initial build process.
These directories and files are fully built (for Mach) in the RVM1.3 directory on the FTP server, so you can see what you should get if there is trouble.
The lib, bin, and include directories are the interface for application build procedures. References to them will get the correct headers, libraries, and utility programs. They are particularly important under Mach since lib and bin are symbolic links to machine dependent files. Under Unix, they are directories with the actual files; no support is provided for compilation for multiple machine types. Also, the build procedure under Unix leaves all derived files in the same directories as the sources since the OBJS directory structure is not built. The structure used by the Mach build process could also be used for Unix by modifying the Makefiles to use the machine specifier used on your system to replace the @sys construct used in Mach.
The Makefiles included are intended to both install the system and help with the building of customized versions. The process can be controlled from two levels: the top level, at the RVM1.3 directory, for builds of all the subsystems. The second level is builds of a specific subsystem from its source directory. From the top level, the following build options are available:
creates directory structures for derived files and then compiles all libraries, utility and test programs. This should be done only once.
does a complete recompilation and installation. If the necessary directories are not present, they will be built. All existing derived files are removed before the build so that a complete rebuild is guaranteed.
builds all subsystems from existing derived files, and copies the interface files to the appropriate directories. Intended for a release after work has been done on one or more subsystems.
builds all subsystems from existing derived files, but does not copy files to the interface directories.
removes all derived files from the subsystem directories and
OBJS, if it is built. The
lib
and
bin
directories are not affected. This is recommended to recover disk
space after the initial build of RVM if you do not plan on making
modifications to RVM.
builds the specified subsystem.
From the subsystem level, the install, reinstall, all, and clean options are also available, and do their respected operation for only that subsystem. Further options are also available to create specific libraries, utilities, and test programs. These can be determined by examining the subsystem Makefiles.
The version of Make used in Mach 2.6 will automatically include
a file called Makeconf if it exists. This is used for common
configuration macros. Since Unix Make doesnt support this, but does
have an include facility, most configuration macros are actually
put in the file RVM1.3/build_defs, which is included are necessary.
The Makeconf files contain Mach specific things only. To allow the
same Makefiles to be used for both operating systems, most
directories are referred to by macro names which will evaluate
appropriately for the system. This is done using only the most
basic Make features in the hope that the build will be as portable
as possible. One artifact of this method is that the installation
will produce a number of error messages followed by messages of the
form
"*** Exit 1 (ignored)"
as the inappropriate
commands fail. Just note that this is expected behavior.
There are a number of other build options that are only used for the Coda File System. These should be ignored since they require many files not included in the RVM distribution.
Under Mach, RVM is internally multi-threaded and expects to be linked with the Cthreads package. Either the Mach threads or coroutine libraries can be used. Coroutines are recommended while debugging since program behavior can be repeatable.
Under Unix, RVM is at this time single threaded. To allow RVM to compile, symbolic links are created for the header file cthreads.h to the file dummy_cthreads.h. No thread library will be required at link time. If you have a thread package that you would like to try to use, the file rvm_lwp.h can be used as a model of how to create a header file for another thread package. This not recommended for the faint of heart!
This distribution of RVM includes two test programs: testrvm, a basic "sanity checker", and rvm_basher, a stress test. You should run these after installing RVM. Running the simple test involves two steps: creating a log file with rvmutl, and then running the test. Its most easily done from the RVM1.3/tests directory, but if you copy the map_*_files and the t*_chk_files to any other place, you can run it there. The procedure is:
.../RVM1.3/bin/rvmutl * init_log log_file 10k * quit .../RVM1.3/bin/testrvm
It will then ask three questions, all of which you should simply enter the default "No" by hitting return. It will do some basic transactions and compare virtual memory with the map_chk_file for correct operation. It will report correct operation. Any other termination is an indication of trouble.
rvm_basher is a much more thorough test and should be run for several hours (we have fixed bugs that would only show up after running it for many days on Pentium 90 computers!). The program requires that a data segment be built as well as a log file. The tests will randomly allocate, modify, and deallocate recoverable memory, periodically stopping to check virtual memory against the data segment. Any discrepancy will be reported and the program will stop. Otherwise it will run until you stop it, given the parameters supplied in this distribution.
rvm_basher can also be used a a sample RVM application since it uses most of the RVM library and also the dynamic allocator, RDS, and segment loader. The setup procedure for running it is similar to what most applications will require, although their heap and static sizes wiil most likely differ.
First, cd to a directory in which you can build about 2M of files. Create a log file using rvmutl as in the the previous test:
.../RVM1.3/bin/rvmutl * init_log basher_log 600k * sizeof page * quit
To prepare a data segment, first load ...RVM1.3/bin/rvm_basher and run the following commands:
.../RVM1.3/bin/rvm_basher * log basher_log * show_break * run
The program will load and print the current break point (with sufficient space added to compensate for internal requirements). This will become the load point for the data segment. Application programs will also have to determine their break point after initialization. If you plan to use pointers in recoverable storage, be sure to add enough to the breakpoint to allow growth of your application. Otherwise you will have to modify the pointers or re-initialize recoverable storage to allow the break point to be moved.
Next, create a data segment with an RDS heap using the rdsinit utility. The static length must be the size of a page on your machine as indicated by rvmutl when you created the log. (Note: all mappings must be integral page sized in RVM; your application can use as many pages as need in both the heap and static regions -- the basher needs only the minimum.)
.../RVM1.3/bin/rdsinit basher_log basher_data Enter the length of the file/device basher_data: 1000000 < rdsinit will zero the file and initialize RVM > starting address of rvm: < something much above the break point > heap len: 0xf0000 static len: < static len > nlists: 100 chunksize: 128
NOTE: On Linux Sparc, you must give page size 8192, not 4096.
You are now ready to run the basher:
.../RVM1.3/bin/rvm_basher < .../RVM1.3/tests/basher_parms
The file .../RVM1.3/tests/basher_parms contains all the parameters to run a thorough test of RVM and its subsystems. As indicated earlier, allowing it to run for at least several hours is recommended.
The benchmark program used for the SOSP 1993 paper, Lightweight Recoverable Virtual Memory ], is included in this distribution. It is not known to work on Unix or on versions of Mach outside the Computer Science Dept. at CMU. Unlike RVM, it is dependent on C++, and may have dependencies on the exact compiler version. The version of C++ used at CMU is AT & T C++ Version It is included as documentation for the methods used to produce the data for the paper and most likely will not work as distributed on your machine.
One known problem is that the CMU CS C++ systems use < libc.h > instead of < stdlib.h > which does not work correctly here. Before attempting to compile the benchmark, you will have to change this, assuming your < stdlib.c > works. The inclusion of < mach.h > , which is required for compilation, will also cause trouble on Unix systems. Also, attempts to use other C++ systems have produced multiple definition errors for some of the standard functions such perror.
Because of these problems, no attempt is made to compile the benchmark. However, a Makefile that works at CMU is included as a starting point if you wish to try building it.
Email should be sent to rvm@cs.cmu.edu. If you send mail to this account when you import RVM, you will be put on a mailing list of future upgrades and bug fixes.
Please understand that if you modify RVM, or any other parts of this distribution, we cannot provide assistance in debugging your modifications.