Comments for Arnaldo's Ramblings

Comment on perf on parisc64 by acmel

acmel — Sat, 09 Jan 2010 12:29:14 +0000

An endianness field, yes, so that we can record at native endianness, i.e. at full speed, and only when decoding perform conversion.

I’ll take a look at exactly how is it that ELF handles this, trying to use some solution that was validated in some other domain already.

Comment on perf on parisc64 by fweisbec

fweisbec — Fri, 08 Jan 2010 22:28:08 +0000

Ah! Yeah that reminds me we need to add an endianness field in the headers, so
that we can cross-handle the bitmap of features correctly.

I’ll send a patchlet for that soon.

Comment on Quickie: dwarves 1.3 released, bitfields edition by fche

fche — Thu, 07 Feb 2008 14:47:07 +0000

BTW, the expensive routine being used is probably not any particular timestamp-gathering function, but the kprobes breakpoint system underneath these probes.

Comment on libdwfl conversion by Diego Flameeyes Pettenò

Diego Flameeyes Pettenò — Mon, 02 Apr 2007 02:30:41 +0000

Okay FWIW libdwfl does not seem to be a proper installed library, it still links to libdw, so it works fine here

Comment on libdwfl conversion by Diego Flameeyes Pettenò

Diego Flameeyes Pettenò — Fri, 30 Mar 2007 22:22:56 +0000

This starts being an issue here on Gentoo as libdwfl does not seem to get installed.
I’ll probably have to continue with an old pahole version for now.

Oh well, will just have to straighten it up a bit

Comment on News fom dwarves land by admin

admin — Sun, 25 Feb 2007 20:24:54 +0000

Ok, today I’ve read a bit more about cmake to get its instruction files in a better shape, including solving the problem that is affecting you, that is, to have elfutils installed in a non-standard path, I’ve tested installing elfutils in ~/usr/local and with the just commited pahole csets it just works, in your case just edit cmake/modules/FindDWARF.cmake replacing ~/usr/local/ with ~/usr/ and you should be set.

CMake is being adopted by more and more projects, notably KDE4, that will provide the impetus for it to be properly supported in all distros, providing competition for SCons and the venerable auto-tools, which is always good.

Comment on News fom dwarves land by admin

admin — Sun, 25 Feb 2007 13:57:38 +0000

build systems, oh well, the whole story is that I never studied autoconf/automake enough to get over the feeling that it was overly complicated, so when I felt the need for using such a system in pahole I made this comment about having to finally learn the autotools and a friend, Helio Chissini, suggested using cmake, he was so enthusiastic about it that quickly provided the initial CMakeFile.txt (camel case, bah, my first annoyance on cmake ) and the FindDWARF.cmake.

But I continue a lazy boy on such build systems… I’m now trying to fix this tho, should have something for you to test later today. Or perhaps I should just get back to using the autotools.

Comment on News fom dwarves land by Jeff Muizelaar

Jeff Muizelaar — Sun, 25 Feb 2007 03:54:23 +0000

The cmake build system is a pain. I used to track pahole development very closely and even wrote a tool based on it for tracing shared libraries. Unfortunately, since the switch to cmake I haven’t been able to build it.

The first problem was that I didn’t have a new enough version of cmake, so I had to compile cmake from source. Now that I have a working cmake, I haven’t been able to get cmake to find my libdwarf headers (they’re in $HOME/usr/include). I’ve tried editing FindDWARF.cmake but that doesn’t seem to work either.

Comment on Unfolding structs by IxLabs » Blog Archive » This case deserves to be a classic (part 1)

IxLabs » Blog Archive » This case deserves to be a classic (part 1) — Sat, 10 Feb 2007 20:34:32 +0000

[...] Back to the slab info. There is one more piece of information which proved to be crucial. In the trace we can actually see the offset were the socket object is modified: 0×34. Now, at that time I did all the math by hand, but thanks to Arnaldo’s pahole we can now put the dwarves to work: [...]

Comment on The 5th dwarf by Frank Ch. Eigler

Frank Ch. Eigler — Tue, 30 Jan 2007 12:10:44 +0000

You may find that if the data extraction variety and robustness needs
start climbing, a more general framework like systemtap will justify
itself. See for example here, which hints at what is required to give
proper per-thread nesting.
http://sourceware.org/systemtap/tutorial/node5.html
Or in-situ data filtering:
http://sourceware.org/systemtap/tutorial/node9.html