ThreadSanitizer under Mac OS X hangs on some Google Chromium tests on which "aspacem" was previously reported.

[GoogleCodeExporter]

To reproduce:
1. Check out and build the Chromium tests (see dev.chromium.org for the
build instruction)
2. Run the current version of ThreadSanitizer on one of the tests from the
list below.
3. Observe the hangs.

At the moment we consider this a Valgrind bug (see
https://bugs.kde.org/show_bug.cgi?id=192634 for more details), because
previously (as of May 2009 -- see the corresponding Valgrind versions)
every Valgrind tool reported the "aspacem" warnings for these tests. After
some refactoring of the address space manager done by Julian Seward in
August 2009 the "aspacem" warnings have gone, but the bug seems to still
remain within ThreadSanitizer.

Valgrind authors have decided to rework the aspacemgr completely in 3.6.0
(see https://bugs.kde.org/show_bug.cgi?id=203254). We hope this should fix
our problems as well.

Original issue reported on code.google.com by ramosian.glider@gmail.com on 14 Sep 2009 at 3:30

[GoogleCodeExporter]

The list of vulnerable tests from net_unittests:
FileStreamTest.*Async*
HostResolverImplTest.*
TestCompletionCallbackTest.Simple
FtpNetworkTransactionTest.*
HttpNetworkLayerTest.GET
HttpNetworkLayerTest.SimpleGET
HttpNetworkTransactionTest.*
ProxyScriptFetcherTest.*
SOCKS5ClientSocketTest.*
SOCKSClientSocketTest.*
TCPClientSocketPoolTest.*
URLRequestTest.*
URLRequestTestHTTP.*
URLRequestTestFTP.*
SSLClientSocketTest.*
DiskCacheTest.*
DiskCacheEntryTest.*Async*
DiskCacheEntryTest.DoomSparseEntry

and from base_unittests:
WorkerPoolTest.PostTask

The "aspacem" warnings can be observed under ThreadSanitizer r1102 (Valgrind 
r10454,
VEX r1908)

Original comment by ramosian.glider@gmail.com on 16 Sep 2009 at 1:26

[GoogleCodeExporter]

A new aspacem reporter in net_unittests is FlipNetworkTransactionTest.Connect

Original comment by ramosian.glider@gmail.com on 13 Oct 2009 at 6:48

[GoogleCodeExporter]

net_unittests again: DiskCacheEntryTest.CancelSparseIO

Original comment by ramosian.glider@gmail.com on 14 Oct 2009 at 5:37

[GoogleCodeExporter]

and once more: FlipNetworkTransactionTest.ResponseWithoutSynReply

Original comment by ramosian.glider@gmail.com on 26 Oct 2009 at 11:44

[GoogleCodeExporter]

This bug conforms to https://bugs.kde.org/show_bug.cgi?id=216837, which turns 
out to
be some VEX problem probably unrelated to address space manager.

ThreadSanitizer built using:
  make -j8 CXXFLAGS+="-DDEBUG=1 -DTS_INSTR_VALGRIND_HERE=1 "
crashes on a debug assertion in VgTidToTsTid(). This could be worked around 
using the
attached patch, but the hack leads to false positives.

Original comment by gli...@google.com on 8 Dec 2009 at 11:09

Attachments:

• tsan.diff

[GoogleCodeExporter]

Alex, would you please add the related reg test to our suite? 

Original comment by konstant...@gmail.com on 20 Jan 2010 at 10:08

[GoogleCodeExporter]

unit_tests: DnsMasterTest.* and URLFetcher*Test.*

Original comment by gli...@chromium.org on 3 Mar 2010 at 12:31

[GoogleCodeExporter]

Kostya, the corresponding test in racecheck_unittest is called
MacTests.DISABLED_WqthreadRegressionTest.

Original comment by gli...@chromium.org on 3 Mar 2010 at 12:32

[GoogleCodeExporter]

A rough fix is available as of r1842.

Original comment by gli...@chromium.org on 15 Mar 2010 at 4:52

[GoogleCodeExporter]

Is it fixed or worked around?
Can we close the issue?

Original comment by timurrrr on 27 May 2010 at 8:18

[GoogleCodeExporter]

Original comment by ramosian.glider@gmail.com on 27 May 2010 at 8:46

• Changed state: Fixed