darkkirb/old-cross-binutils
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
old-cross-binutils/sim/ppc/RUN

889 lines
25 KiB
Text
Raw Normal View History

Update to rev 1.0 from andrew
1996-02-16 15:51:57 +00:00
PSIM - model the PowerPC environment
Copyright (C) 1994-1996, Andrew Cagney <cagney@highland.com.au>.
----------------------------------------------------------------------
Running PSIM
This file describes how to run the program PSIM.
o Walk through a number of examples from the
pre-built tar archive psim-test.
o Looks at the device tree used by PSIM.
o Notes on building a programmer environment to
use with PSIM (BSD/UEA and BUG/OEA)
----------------------------------------------------------------------
RUNNING PSIM:
The compressed tar archive psim-test available from:
ftp://ftp.ci.com.au/pub/psim/psim-test-1.0.tar.gz
or ftp://cambridge.cygnus.com/pub/psim/psim-test-1.0.tar.gz
contains a number of pre-built programs for running under PSIM. Each
pre-built binary is built both big and little endian. The suffixes
.be/.le (executables) .bo/.lo (object files) and .ba/.la (libraries)
are used.
To run one of these programs, use:
powerpc-unknown-eabi-run <image>
vis
powerpc-unknown-eabi-run psim-test/uea/envp
(The program envp prints out your shells environment - very useful
:-). More generally psim is run as (this is part of the output from
the -h option):
psim [ <psim-option> ... ] <image> [ <image-arg> ... ]
Where
<image> Name of the PowerPC program to run.
This can either be a PowerPC binary or
a text file containing a device tree
specification.
PSIM will attempt to determine from the
specified <image> the intended emulation
environment.
If PSIM gets it wrong, the emulation
environment can be specified using the
`-e' option (described below).
<image-arg> Argument to be passed to <image>
These arguments will be passed to
<image> (as standard C argv, argc)
when <image> is started.
<psim-option> See below
The following are valid <psim-option>s:
-m <model> Specify the processor to model (604)
Selects the processor to use when
modeling execution units. Includes:
604, 603 and 603e
-e <os-emul> specify an OS or platform to model
Can be any of the following:
bug - OEA + MOTO BUG ROM calls
netbsd - UEA + NetBSD system calls
chirp - OEA + a few OpenBoot calls
-i Print instruction counting statistics
-I Print execution unit statistics
-r <size> Set RAM size in bytes (OEA environments)
-t [!]<trace> Enable (disable) <trace> option
-o <spec> add device <spec> to the device tree
-h -? -H give more detailed usage
The `-H' option gives a long usage output. This includes a complete
list of all the pre-configured devices.
----------------------------------------------------------------------
RUNNING GDB:
If you built PSIM with gdb then the following is a quick start
tutorial.
At present GDB, if configured big-endian (say) unlike PSIM, does not
support the debugging of little endian binaries. If you find that
your program won't run at all, make certain that GDB and your
program's endianness match.
The most important thing is that before you can run the simulator you
must enable it. For the simulator, gdb is started like any program:
$ powerpc-unknown-eabi-gdb psim-test/uea/envp.be
Next the simulator is enabled. The command `target sim' accepts the
same options as can be specified on the PSIM command line.
(gdb) target sim
To trace the communication between psim and gdb specify `target sim -t
gdb'. Once enabled, the binary needs to be loaded, any breakpoints of
interest set, and the program run:
(gdb) load
(gdb) break main
(gdb) run
.
.
.
In addition, if you are wanting to run a program described by a device
tree you can `attach' to the simulation using (I assume that you have
applied the attach patch):
$ cd psim-test/tree
$ powerpc-unknown-eabi-gdb
(gdb) target sim
(gdb) attach device-tree
(gdb) run
Here GDB takes the programs initial state from the attached
device-tree instead of forcing initialisation.
----------------------------------------------------------------------
PROFILING:
PSIM includes a number of performance monitoring (profiling)
facilities:
o instruction frequency counting
o execution unit modeling (records
effective usage of units).
o instruction cache performance
As discussed in the file INSTALL, each can be configured to individual
requirements.
-i Enable instruction counting.
The frequency of all instructions is tabulated. In
addition (f configured) the hit/miss rate of the
instruction cache is output.
-I Enable execution unit analysis.
In addition to counting basic instructions also model
the performance of the processors execution units
-m <processor>
Select the processor to be modelled.
For execution unit analysis specify the processor that
is to be analysed. By default the 604 is modelled
however, support for other processors such as the
603 and 603e is included.
The output from a performance run (on a P90) for the program
psim-test/profile/bench is below. In this run psim was fairly
agressively configured (see the file INSTALL for compile time
configuration).
CPU #1 executed 41,994 AND instructions.
CPU #1 executed 519,785 AND Immediate instructions.
CPU #1 executed 680,058 Add instructions.
CPU #1 executed 41,994 Add Extended instructions.
CPU #1 executed 921,916 Add Immediate instructions.
CPU #1 executed 221,199 Add Immediate Carrying instructions.
CPU #1 executed 943,823 Add Immediate Shifted instructions.
CPU #1 executed 471,909 Add to Zero Extended instructions.
CPU #1 executed 571,915 Branch instructions.
CPU #1 executed 1,992,403 Branch Conditional instructions.
CPU #1 executed 571,910 Branch Conditional to Link Register instructions.
CPU #1 executed 320,431 Compare instructions.
CPU #1 executed 471,911 Compare Immediate instructions.
CPU #1 executed 145,867 Compare Logical instructions.
CPU #1 executed 442,414 Compare Logical Immediate instructions.
CPU #1 executed 1 Condition Register XOR instruction.
CPU #1 executed 103,873 Divide Word instructions.
CPU #1 executed 104,275 Divide Word Unsigned instructions.
CPU #1 executed 132,510 Extend Sign Byte instructions.
CPU #1 executed 178,895 Extend Sign Half Word instructions.
CPU #1 executed 871,920 Load Word and Zero instructions.
CPU #1 executed 41,994 Move From Condition Register instructions.
CPU #1 executed 100,005 Move from Special Purpose Register instructions.
CPU #1 executed 100,002 Move to Special Purpose Register instructions.
CPU #1 executed 804,619 Multiply Low Word instructions.
CPU #1 executed 421,201 OR instructions.
CPU #1 executed 471,910 OR Immediate instructions.
CPU #1 executed 1,292,020 Rotate Left Word Immediate then AND with Mask instructions.
CPU #1 executed 663,613 Shift Left Word instructions.
CPU #1 executed 1,151,564 Shift Right Algebraic Word Immediate instructions.
CPU #1 executed 871,922 Store Word instructions.
CPU #1 executed 100,004 Store Word with Update instructions.
CPU #1 executed 887,804 Subtract From instructions.
CPU #1 executed 83,988 Subtract From Immediate Carrying instructions.
CPU #1 executed 1 System Call instruction.
CPU #1 executed 207,746 XOR instructions.
CPU #1 executed 23,740,856 cycles.
CPU #1 executed 10,242,780 stalls waiting for data.
CPU #1 executed 1 stall waiting for a function unit.
CPU #1 executed 1 stall waiting for serialization.
CPU #1 executed 1,757,900 times a write-back slot was unavailable.
CPU #1 executed 1,088,135 branches.
CPU #1 executed 2,048,093 conditional branches fell through.
CPU #1 executed 1,088,135 successful branch predictions.
CPU #1 executed 904,268 unsuccessful branch predictions.
CPU #1 executed 742,557 branch if the condition is FALSE conditional branches.
CPU #1 executed 1,249,846 branch if the condition is TRUE conditional branches.
CPU #1 executed 571,910 branch always conditional branches.
CPU #1 executed 9,493,653 1st single cycle integer functional unit instructions.
CPU #1 executed 1,220,900 2nd single cycle integer functional unit instructions.
CPU #1 executed 1,254,768 multiple cycle integer functional unit instructions.
CPU #1 executed 1,843,846 load/store functional unit instructions.
CPU #1 executed 3,136,229 branch functional unit instructions.
CPU #1 executed 16,949,396 instructions that were accounted for in timing info.
CPU #1 executed 871,920 data reads.
CPU #1 executed 971,926 data writes.
CPU #1 executed 221 icache misses.
CPU #1 executed 16,949,396 instructions in total.
Simulator speed was 250,731 instructions/second
----------------------------------------------------------------------
PSIM CONFIGURATION - THE DEVICE TREE
Internally PSIM's configuration is controlled by a tree data
structure. This structure, created at run-time, intentionally
resembles the device tree used by OpenBoot firmware to describe a
machines hardware configuration.
PSIM can either create its device tree using a builtin emulation or
from one read in from a file.
During startup, the device tree is created using the following steps:
o Initial empty tree is created
o Any tree entry options specified on the
command line are merged in (the -o <entry>
option is used).
It should be pointed out that most of the
command line options (eg -r, -e, -m, -t
are all just short hand for corresponding
-o options).
o If the specified program is a device tree spec, that
is loaded.
If the specified program is a text file it is assumed
that that file contains a further specification of the
simulators device tree. That tree is loaded and
merged with the current tree options.
o The selected emulation fills out any remaining details.
By this stage the emulation environment that the program
needs will either be specified in the device tree
(through the -e option) or determined from the
characteristics of the binary.
The selected emulation will then fill out any missing
nodes in the device tree.
Most importantly earlier additions to the tree are not overridden by
later additions. Thus, command line options override information
found in the program file and both override any emulation entries.
The following is a summary of the most useful runtime configuration
options:
-e <os-emul>
-o '/openprom/options/os-emul <os-emul>'
Run program using the <emulation> run-time
environment.
-r <ram-size>
-o '/openprom/options/oea-memory-size <ram-size>'
Set the size of the first bank of memory
(RAM from address 0 up).
-t print-device-tree
-o '/openprom/trace/print-device-tree 1'
-t dump-device-tree
-o '/openprom/trace/dump-device-tree 1'
Print out the device tree once it has been fully
populated. For dump-device-tree, exit simulator after
dumping the tree.
PSIM is able to reload the dumped device tree.
The format of the dumped tree is under development.
-o '/openprom/options/smp <N>'
Enable <N> processors for the simulation run.
See the directory psim-test/oea for an example.
-o '/openprom/options/alignment <N>'
Where <N> is 1 - nonstrict or 2 - strict.
Specify if the missaligned access are allowed
(non-strict) or result in an alignment exception
(strict).
Devices (if included in the file device_table.c) can also be specified
in a similar way. For instance, to add a second serial port, a
command like:
-o '/iobus@0x400000/console@0x000010'
would create a `console' device at offset 0x10 within the `iobus' at
memory address 0x400000.
For more detailed information on device specifiers see the notes on
the function dump_device_tree in the file device.c (found in the
source code).
----------------------------------------------------------------------
BUILDING A BUG/OEA DEVELOPMENT ENVIRONMENT
Background:
Included in many PowerPC systems is Motorola's BUG monitor. This
monitor includes, for client programs, a set of services that allow
that program to interact with hardware devices such as the console using
a simple system call interface.
PSIM is able to emulate a number of the services (including the
console IO calls). If additional services are needed they can easily
be added.
Cygnus support's newlib library includes includes an interface to the
MOTO BUG services. The notes below discuss how I both built and run
programs compiled using this library on PSIM.
The only confusing part about building a development environment based
around newlib/binutils/gcc is a chicken/egg problem with include
files:
For GCC to build, a fairly complete set of include
files must be installed but newlib won't install its
include files until it has been built with gcc ...
I get around this by installing the problematic include files by hand.
Preparation:
The following files are needed:
From your favorite FTP site, the sources to gas/ld and gcc - mine
happens to be archie.au :
ftp://archie.au/gnu/binutils-2.6.tar.gz
ftp://archie.au/gnu/gcc-2.6.2.tar.gz
From ftp://ftp.cygnus.com/pub/newlib the source code to a library:
ftp://ftp.cygnus.com/pub/newlib/newlib-1.7.0.tar.gz
From ftp://ftp.ci.com.au/pub/psim some minor patches and updates to
the above library:
ftp://ftp.ci.com.au/pub/psim/newlib-1.7.0+float+ppc-asm.tar.gz
ftp://ftp.ci.com.au/pub/psim/newlib-1.7.0+ppc-fix.diff.gz
ftp://ftp.ci.com.au/pub/psim/binutils-2.6+note.diff.gz
In addition you'll need to decide where you will be installing the
development environment. You will notice that in the below I install
things well away /usr/local instead installing everything under its
own directory in /applications.
Method:
These notes are based on an installation performed on a Sun-OS-4/SPARC
host. For other hosts and other configurations, the below should be
considered as a guideline only.
o Sanity check
$ cd .../scratch # your scratch directory
$ ls -1
binutils-2.6.tar.gz
binutils-2.6+note.diff.gz
gcc-2.7.2,tar.gz
newlib-1.7.0+float+ppc-asm.tar.gz
newlib-1.7.0+ppc-fix.diff.gz
newlib-1.7.0.tar.gz
o Unpack/build/install binutils
This is done first so that there is a gas/ld ready
for the building of GCC and NEWLIB.
$ cd .../scratch
$ gunzip < binutils-2.6.tar.gz | tar xf -
$ cd binutils-2.6
Optionally apply the note patch
$ gunzip ../binutils-2.6+note.diff.gz | patch
Then continue with the build
$ ./configure --target=powerpc-unknown-eabi \
--prefix=/applications/psim
$ make
$ make install
$ cd ..
$ rm -rf binutils-2.6
This also creates much of the installation directory
tree.
o Unpack newlib, install the include files so that they
are ready for GCC's build.
$ cd .../scratch
$ gunzip < newlib-1.7.0.tar.gz | tar xf -
New lib-1.7.0 had a few minor bugs (fixed in current):
the header files float.h and ppc-asm.h were missing;
the configure and Makefile's for the rs6000 (ppc) directory
contained typos:
$ cd .../scratch
$ cd newlib-1.7.0
$ gunzip < ../newlib-1.7.0+float+ppc-asm.tar.gz | tar xvf -
$ gunzip < ../newlib-1.7.0+ppc-fix.diff.gz | patch -p1
Finally copy the include files to where GCC will see them:
$ cd .../scratch
$ cd newlib-1.7.0/newlib/libc
$ tar cf - include | \
( cd /applications/psim/powerpc-unknown-eabi && tar xf - )
o Unpack/build gcc
$ cd .../scratch
$ gunzip < gcc-2.7.2,tar.gz | tar xf -
$ cd gcc-2.7.2
$ ./configure --target=powerpc-unknown-eabi \
--prefix=/applications/psim
$ make
$ make install
$ cd ..
$ rm -rf gcc-2.7.2
Gcc likes to install its own dummy version of float that
just returns an error.
$ more /applications/psim/lib/gcc-lib/powerpc-unknown-eabi/2.7.2/include/float.h
$ rm /applications/psim/lib/gcc-lib/powerpc-unknown-eabi/2.7.2/include/float.h
o Finish building/installing newlib
$ cd .../scratch
$ cd newlib-1.7.0
$ ./configure --target=powerpc-unknown-eabi \
--prefix=/applications/psim
Your path will need to include the recently installed
gas/gcc when building. Either add it to your path or
use:
$ PATH=/applications/psim/bin:$PATH make
$ PATH=/applications/psim/bin:$PATH make install
o Finally, test out the build
$ cat hello.c
main()
{
printf("hello world\n");
}
The binary is linked with an entry point less than 0x100000
(1mb) so that psim will recognize the binary as needing
the BUG/OEA instead of the BSD/UEA runtime environment.
$ powerpc-unknown-eabi-gcc -v -o hello \
-Wl,-Ttext,0x4000,-Tdata,0x10000 \
/applications/psim/powerpc-unknown-eabi/lib/mvme-crt0.o \
hello.c \
-lc -lmvme
$ powerpc-unknown-eabi-objdump -h hello
$ powerpc-unknown-eabi-run hello
It is also possible to force psim to use a specific
run-time environment using the -e option vis:
$ powerpc-unknown-eabi-run -e bug hello
----------------------------------------------------------------------
BUILDING A BSD/UEA DEVELOPMENT ENVIRONMENT
Background:
For a UEA to be useful it needs a supporting run-time environment.
PSIM implements a runtime environment based on the NetBSD system call
interface.
More than any thing, this user level emulation was the first
implemented because I happened to have the NetBSD source code lying
lying around.
Preparation:
This requires the NetBSD-1.1 source tree online. It can either be
obtained vi ftp:
try http://www.netbsd.org or ftp://ftp.netbsd.org
Alternatively obtain one of the NetBSD cdrom's. Patches to this source
tree that fill out much of the PowerPC code are available in:
ftp://ftp.ci.com.au/pub/clayton
Fetch everything in that directory - diffs, tar archives and scripts.
In addition a patch to binutils is in:
ftp://ftp.ci.com.au/pub/psim/binutils-2.6+note.diff.gz
Finally you'll require a compiler and assembler/linker:
gcc-2.7.2.tar.gz
binutils-2.6.tar.gz
Method:
These notes are based on an installation performed on a Solaris2/x86
host. For other hosts and other configurations, the below should be
considered as a guideline only.
o Sanity check
I assume that you have already obtained the NetBSD-1.1 source
code and unpacked it into the directory bsd-src. While the
full NetBSD source tree may not be needed, things are easier
if it is all online.
$ cd .../scratch
$ ls -1
binutils-2.6.tar.gz
binutils-2.6.tar.gz
clayton-include-960203.diff.gz
clayton-lib-960203.diff.gz
clayton-lib-960203.tar.gz
clayton-sys-960203.diff.gz
clayton-sys-960203.tar.gz
clayton-utils-960203.tar.gz
clayton.chown.sh
clayton.install.sh
clayton.lorder.sh
clayton.make.sh
clayton.usr.bin.make.diff
gcc-2.7.2.tar.gz
gcc-2.7.2+sys-types.diff.gz
o Unpack the bsd source code (if you haven't already)
$ cd .../scratch
$ mkdir bsd-src
$ cd bsd-src
$ for d in /cdrom/bsdisc_12_95_disc2/NetBSD-1.1/source/*11
do
echo $d
cat $d/*.?? | gunzip | tar xf -
done
Flatten the directory structure a little.
$ mv usr/src/* .
$ rmdir usr/src usr
$ cd ..
o Unpack/build/install binutils
$ cd .../scratch
$ gunzip < binutils-2.6.tar.gz | tar xf -
$ cd binutils-2.6
Optionally apply the note patch
$ gunzip ../binutils-2.6+note.diff.gz | patch
Then continue with the build
$ ./configure --target=powerpc-unknown-eabi \
--prefix=/applications/psim
$ make
$ make install
$ cd ..
$ rm -rf binutils-2.6
This has the intended side effect of partially populating
the psim directory tree which makes follow on steps easier.
o Fill out the install directory with a few additions (if
install -d works, this can be simplified).
$ mkdir \
/applications/psim/bsd-root \
/applications/psim/bsd-root/usr \
/applications/psim/bsd-root/usr/share \
/applications/psim/bsd-root/usr/share/doc \
/applications/psim/bsd-root/usr/share/doc/psd \
/applications/psim/bsd-root/usr/share/doc/psd/19.curses \
/applications/psim/bsd-root/usr/include \
/applications/psim/bsd-root/usr/lib \
o Make the bsd and gnu include directories point to the same
location.
GCC expects include files to be in one location while the
bsd install expects them in a second. The link is in
the direction below because bsd's install also insists on
a directory (not a link) for its install destination.
$ ln -s ../bsd-root/usr/include \
/applications/psim/powerpc-unknown-eabi/include
o Build/install Berkeley make
In building Berkeley make from the NetBSD-1.1 source tree
a number of problems may be encountered.
These problems have been fixed in NetBSD-current (after
4/2/96 (ie start Feb)) you should probably obtain that
version of make. Alternatively, you can try following the
notes below that got make working on a Solaris-2.5/x86
host.
$ cd .../scratch
$ cd bsd-src/usr.bin/make
$ pwd
.../scratch/bsd-src/usr.bin/make
Copy/stub some additional include files that your host may not
have.
$ cp ../../include/ranlib.h ranlib.h
$ mkdir sys
$ cp ../../sys/sys/cdefs.h sys/cdefs.h
$ mkdir machine
$ touch machine/cdefs.h
Edit/fix some of the BSDisms. The patch file indicated
contains fixes I found when compiling on my host, your
host will probably differ.
$ gunzip < ../../../clayton.usr.bin.make.diff.gz | patch
Build it with some extra flags (again your flags will differ):
$ make -f Makefile.boot CC='gcc -g -DPOSIX'
With bmake built, install it into the target specific bin
directory:
$ cp bmake /applications/psim/powerpc-unknown-eabi/bin/make
$ cd ../../..
o Patch/install Berkeley make's include (mk) files.
$ cd .../share
$ cd bsd-src/share
$ tar cf - mk | ( cd /applications/psim/bsd-root/usr/share \
&& tar xf - )
$ cd ../..
o Set up a number of wrapper scripts for bmake so that it works.
In addition to needing BSD make the build process assumes
a number of BSD specific commands. To get around this
several wrapper scripts are available.
powerpc-unknown-eabi-make (clayton.make)
Front end to Berkeley make setting it up for a
cross compilation
/applications/psim/bin/powerpc-unknown-eabi-make
chown (clayton.chown)
Wrapper that does not do any thing.
Avoids the need to be root when installing.
/applications/psim/powerpc-unknown-eabi/bin
install (clayton.install)
Wrapper to strip away a number of bsd specific install
arguments.
/applications/psim/powerpc-unknown-eabi/bin/install
lorder (clayton.lorder)
Tweaked lorder script that will use nm etc from
binutils.
/applications/psim/powerpc-unknown-eabi/bin/lorder
o Apply the remaining patches for the BSD source tree
$ cd .../scratch
$ cd bsd-src
Diffs are applied using something like:
$ gunzip < ../clayton-include-960203.diff.gz | more
...
The patch to sys/dev/pci/ncr.c.rej might fail.
The tar archives have a different problem, you need
to remove the `src' prefix. I used
$ ln -s . src
$ gunzip < ../clayton-lib-960203.tar.gz | tar xtf -
...
So that src/xxx unpacked into ./xxx
$ cd ..
o Install the include files
$ cd .../scratch
$ cd bsd-src/include
$ powerpc-unknown-eabi-make install
$ cd ../..
o Install a few other include files.
As with building libnew, the bsd build process has
several include file problems.
$ cd .../scratch
$ cd bsd-src
$ cp gnu/lib/libg++/g++-include/values.h \
/applications/psim/powerpc-unknown-eabi/include
$ cp lib/libcurses/curses.h \
/applications/psim/powerpc-unknown-eabi/include
$ cd ..
o Build/install gcc
$ cd .../scratch
$ gunzip < gcc-2.7.2,tar.gz | tar xf -
$ cd gcc-2.7.2
GCC and BSD (for PowerPC) have a conflicting type
declaration. The patch below gets around this
problem:
$ gunzip < ../gcc-2.7.2+sys-types.diff.gz | patch
Other than that, assuming the include files installed
okay, the rest should be fine ....
$ ./configure --target=powerpc-unknown-eabi \
--prefix=/applications/psim
$ make CC=gcc
$ make CC=gcc install
$ cd ..
$ rm -rf gcc-2.7.2
o Build/install the Berkeley library:
$ cd .../scratch
$ cd bsd-src/lib
$ powerpc-unknown-eabi-make
$ powerpc-unknown-eabi-make install
$ cd ../..
If you encounter problems try the following: an include
file not yet installed; a directory not yet created;
running the hosts version of a program instead of a
bsd version.
o Build/run a simple BSD program
$ cd .../scratch
$ cd bsd-src/usr.bin/printenv
$ powerpc-unknown-eabi-make
$ powerpc-unknown-eabi-run printenv
.
.
.
----------------------------------------------------------------------
Reference in a new issue Copy permalink