GNU/Linux |
CentOS 2.1AS(Slurm) |
|
![]() |
njamdpm(1) |
![]() |
njamdpm − Not Just Another Malloc Debugger Post-Mortem
njamdpm [OPTIONS] <HEAP FILE>
njamdpm is a companion utility that allows you to examine the persistent heap saved by libnjamd(3) You can do things like query for certain addresses, show memory leaks, and show all past allocated memory. As of NJAMD 0.6.0, gdb(1) is required to make sense of the return addresses.
Options
HEAP FILE
The heap file will be in the current directory with a name of the form njamd-<pid>-heap, but only if NJAMD_PERSISTANT_HEAP was in the environment at the time of program execution
−a address
Search through the heap file for a chunk of memory that contains address. This can be VERY helpful when using gdb. Simply find the address that you accessed to cause the segmentation fault, use njamdpm to look it up in the heap, and viola! You have all sorts of info about the chunk: When it was allocated, when it was freed, how big is is, etc.
−d depth
When displaying return address info, only display depth return addresses. The max is specified in ./include/lib/njamd.h in the define TRACE_DEPTH (default is 3).
−t |
Trim the heap file down to only the used portion. This is useful if for some reason the program somehow exits without trimming its own heap file down first. Note that when the heap file appears huge it’s not actually taking up disk space. | ||
−s |
Dump basic status info about peak memory usage, NJAMD overhead, etc. Useful for determining if you should buy more ram, or write me an angry email :) | ||
−l |
Dump memory leaks in the heap. Also shows you info about where the memory was leaked, along with a total. Do note that this total and the subtotals are aligned bytes. They are aligned to the alignment of your architecture, or as specified by the value the NJAMD_ALIGN environment variable had when the heap was created. | ||
−f |
Dump freed memory in the heap. This option is only available if LIBNJAMD ran without NJAMD_CHK_FREE=none set. |
Using gdb
with njamdpm
When a segmentation fault happens, it’s because, of
course, you accessed an invalid address. So all you need to
do is get gdb to give you the address you accessed, and then
feed it to njamdpm. Ie if the segfault occurs on a line that
does buf[i] = 2, issue print &buf[i] to gdb. Note
that libnjamd(3) now has a function __nj_ptr_info
that can be called from gdb that performs all this without
njamdpm.
To get gdb to translate these return addresses into something meaningful, issue
info line *0xaddress
to obtain the line number of the allocation request, or
list *0xaddress
to see the adjacent code as well.
Eventually I hope to add symbol translation right into njamdpm.
Mike Perry <mikepery@fscked.org>
http://freshmeat.net/appindex/development/debugging.html
njamd(3), efence(3), malloc(3), mmap(2), mprotect(2)
![]() |
njamdpm(1) | ![]() |