Re: [Libunwind-devel] Backtrace performance on x86-64 based on Dwarf info

[Milian Wolff]

On Wednesday 28 May 2014 11:08:19 Arun Sharma wrote:

> On Wed, May 28, 2014 at 5:40 AM, Milian Wolff <address@hidden> wrote:

> > Profiling my tracer with perf shows tons of time spent in

> > _ULx86_64_dwarf_find_proc_info and below. Esp. the syscalls related to

>

> > sigprocmask are a hotspot:

> Are you compiling with libatomicops installed?

 

$ grep -A1 ops config.log

configure:15937: checking atomic_ops.h usability

configure:15937: gcc -c -g -O3 -flto conftest.c >&5

--

configure:15937: checking atomic_ops.h presence

configure:15937: gcc -E conftest.c

--

configure:15937: checking for atomic_ops.h

configure:15937: result: yes

--

configure:16981: checking whether to block signals during mutex ops

configure:16995: result: yes

--

ac_cv_header_atomic_ops_h=yes

ac_cv_header_byteswap_h=yes

 

 

So... yes? But what does that have to do with the 0-IP I encounter at the end of a backtrace?

 

Bye

--

Milian Wolff

address@hidden

http://milianw.de