If traces alone are not sufficient, you‘ll need to inspect what gem5 is doing in detail using a debugger (e.g., gdb). You definitely want to use the gem5.debug
binary if you reach this point. Ideally, looking at traces should at least allow you to narrow down the range of cycles in which you think something is going wrong. The fastest way to reach that point is to use a DebugEvent
, which goes on gem5’s event queue and forces entry into the debugger when the specified cycle is reached by sending the process a SIGTRAP
signal. You'll need to to start gem5 under the debugger or have the debugger attached to the gem5 process for this to work.
You can create one or more DebugEvents when you invoke gem5 using the --debug-break=100
parameter. You can also create new DebugEvents from the debugger prompt using the schedBreak()
function. The following example session illustrates both of these approaches:
% gdb m5/build/<ISA>/gem5.debug GNU gdb 6.1 Copyright 2002 Free Software Foundation, Inc. [...] (gdb) run --debug-break=2000 configs/run.py Starting program: /z/stever/bk/m5/build/<ISA>/gem5.debug --debug-break=2000 configs/run.py M5 Simulator System [...] warn: Entering event queue @ 0. Starting simulation... Program received signal SIGTRAP, Trace/breakpoint trap. 0xffffe002 in ?? () (gdb) p curTick $1 = 2000 (gdb) c Continuing. (gdb) call schedBreak(3000) (gdb) c Continuing. Program received signal SIGTRAP, Trace/breakpoint trap. 0xffffe002 in ?? () (gdb) p _curTick $3 = 3000 (gdb)
gem5 includes a number of functions specifically intended to be called from the debugger (e.g., using the gdb call
command, as in the schedBreak()
example above). Many of these are “dump” functions which display internal simulator data structures. For example, eventq_dump()
displays the events scheduled on the main event queue. Most of the other dump functions are associated with particular objects, such as the instruction queue and the ROB in the detailed CPU model. These include:
Function | Effect |
---|---|
schedBreak(<tick>) | Schedule a SIGTRAP to occur at <tick> |
setDebugFlag("<flag>") | Enable a debug flag from the debugger |
clearDebugFlag("<flag>") | Disable a debug flags from the debugger |
eventqDump() | Print out all events on the event queue |
takeCheckpoint(<tick>) | Create a checkpoint at cycle <tick> |
SimObject::find("system.qualified.name") | Returns the pointer to the object with the specified name |
You can debug configuration scripts with the Python debug (PDB) just as you would other Python scripts. You can enter PDB before your configuration script is executed by giving the --pdb
argument to the gem5 binary. Another approach is to put the following line in your configuration script (e.g., fs.py
or se.py
) wherever you would like to enter the debugger:
import pdb; pdb.set_trace()
Note that the Python files under src
are compiled in to the gem5 binary, so you must rebuild the binary if you add this line (or make other changes) in these files. Alternatively, you can set the M5_OVERRIDE_PY_SOURCE
environment variable to “true” (see src/python/importer.py
).
See the official PDB documentation for more details on using PDB.
Valgrind is a dynamic analysis tool used (primarily) to profile a target application and detect the source of run-time errors, as well as detect memory leaks.
For Valgrind to function, the target gem5 binary must have been compiled to include debugging information. Therefore, the gem5.debug
binaries must be used.
To run a check using Valgrind, execute the following:
valgrind --leak-check=yes --suppressions=util/valgrind-suppressions build/{Target ISA}/gem5.debug {gem5 arguments}
The above will run the gem5 and do two things:
The util/valgrind-suppressions
file contains a set of warnings that are reported by Valgrind but are not considered a problem by gem5 developers. Valgrind is known to provide false positives. util/valgrind-suppressions
should be updated as these false positives are revealed. More information about suppressing Valgrind warnings can be found in the Valgrind User Manual.
If a run-time error is received, Valgrind will return an output which looks like the following (taken from the Valgrind Quick Start Guide):
==19182== Invalid write of size 4 ==19182== at 0x804838F: f (example.c:6) ==19182== by 0x80483AB: main (example.c:11)
In this output:
Invalid write
is what kind of error.example.c
. This line is contained within function f
which was called by the main
method at line 11 (also in example.c
).0x804838F
is the code address. This is usually not important.Valgrind may also return warnings about memory leaks, such as:
==19182== 40 bytes in 1 blocks are definitely lost in loss record 1 of 1 ==19182== at 0x1B8FF5CD: malloc (vg_replace_malloc.c:130) ==19182== by 0x8048385: f (a.c:5) ==19182== by 0x80483AB: main (a.c:11)
The stack trace will tell you where the memory leak occurred. If Valgrind states that a block of memory was “definitely lost” then there is a memory leak. However, if Valgrind states that a block was “probably lost”, Valgrind has reason to believe memory is leaking but perhaps not (this is normally if the code is doing something complex with pointers).
If Valgrind returns an output in which a root cause is difficult to determine, try running Valgrind with --track-origins=yes
. This will increase execution time but will provide more information.
The Valgrind User Manual should be consulted for more advanced features.