Re: [DynInst_API:] TLS-based trampguard patch

[Gerard <nouboh@xxxxxxxxx>]

Yes, I'm quite sure the process hangs. If I check the process status at /proc I see: t (tracing stop).

Gerard

2015-02-11 18:02 GMT+01:00 Bill Williams <bill@xxxxxxxxxxx>:

On 02/11/2015 08:06 AM, Gerard wrote:

I'm not sure if this patch should have solved my problem with the
maximum number of threads but with this version I still have the same
problem. If I don't increase the MAX_THREADS constant my instrumented
process still hangs after creating around 40 threads.

Tested on 64bits Gentoo Linux with kernel version 3.18.3-gentoo.

Huh. Didn't expect that the vestigial bits of thread index code would cause a hang; you're sure this is a hang and not a crash mutatee-side?

Anyhow, I've got a more comprehensive patch in the works that eliminates MAX_THREADS and the thread index code entirely, and I'll be sending that one out later.

Gerard

2015-02-06 21:54 GMT+01:00 Bill Williams <bill@xxxxxxxxxxx
<mailto:bill@xxxxxxxxxxx>>:

  As per discussion earlier this week, here's an initial
  implementation of TLS-based trampguards. It is not yet pushed to
  master; I've tested it locally but it has not been through serious
  cross-platform validation yet. Treat as experimental.

  This should (mostly) obviate our hard-coded thread limits, and
  should produce modestly better performance in multithreaded code.
  These changes will not work with an old dyninst runtime library at
  present; that's backwards compatibility that I can wedge back in if
  anyone particularly cares. As always, feedback (both good and bad)
  is encouraged.

  --bw

  Bill Williams
  Paradyn Project
  bill@xxxxxxxxxxx <mailto:bill@xxxxxxxxxxx>

  _______________________________________________
  Dyninst-api mailing list
  Dyninst-api@xxxxxxxxxxx <mailto:Dyninst-api@xxxxxxxxedu>
  https://lists.cs.wisc.edu/mailman/listinfo/dyninst-api

--

--bw

Bill Williams
Paradyn Project
bill@xxxxxxxxxxx