[Bug nptl/13433] New: pthread_create allocates big chunks of memory when launching several threads simultaneously (race condition?)

[sources at shustrik dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=13433

             Bug #: 13433
           Summary: pthread_create allocates big chunks of memory when
                    launching several threads simultaneously (race
                    condition?)
           Product: glibc
           Version: 2.11
            Status: NEW
          Severity: normal
          Priority: P2
         Component: nptl
        AssignedTo: drepper.fsp@gmail.com
        ReportedBy: sources@shustrik.com
    Classification: Unclassified


Created attachment 6074
  --> http://sourceware.org/bugzilla/attachment.cgi?id=6074
C++ code that simulates the problem

First of all, this is not related to the stack size setting.

The behaviour that I'm experiencing is that pthread_create allocates 64Mb
chunks of memory (seen as [anon] in pmap) when multiple threads are being
started simultaneously. This doesn't happen, if I add a slight delay between
thread starts, and threads are created sequentially. At first I blamed boost,
but then I discovered it happens without boost as well, with direct
pthread_create calls. The attached code can be used to show the problem (C++,
requires boost, let me know if You need me to upload a C version).

I get the following output:

~> ./test 2 boost 0
Thread count 2 - Delay 0 - Boost - memory used: 93.5 (1.1) Mbytes
~> ./test 2 pthread 0
Thread count 2 - Delay 0 - Pthread - memory used: 93.5 (1.1) Mbytes
~> ./test 2 boost 10
Thread count 2 - Delay 10 - Boost - memory used: 29.5 (1.1) Mbytes
~> ./test 2 pthread 10
Thread count 2 - Delay 10 - Pthread - memory used: 93.5 (1.1) Mbytes
~> ./test 2 pthread 10
Thread count 2 - Delay 10 - Pthread - memory used: 93.5 (1.1) Mbytes
~> ./test 2 pthread 10
Thread count 2 - Delay 10 - Pthread - memory used: 29.5 (1.1) Mbytes

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.