data race in std::locale::classic()

• Key: STDCXX-523
• Type: Bug
• Status: Open
• Priority: Minor
• Assignee: Martin Sebor
• Reporter: Martin Sebor
• Component/s: 22. Localization
• Affects Version/s: 4.1.3, 4.1.4, 4.2.0
• Fix Version/s: 4.3.0

Description

Intel Thread Checker reports the following data race in std::locale::classic():

$ tcheck_cl ./22.locale.statics.mt --no-global --nthreads=2
Intel(R) Thread Checker 3.1 command line instrumentation driver (24400)
Copyright (c) 2007 Intel Corporation. All rights reserved.
Building project

Running: /build/sebor/stdcxx-trunk-gcc-3.4.6_3-15D/tests/22.locale.statics.mt --no-global --nthreads=2

1. INFO (S1) (10 lines):
2. TEXT:
3. COMPILER: gcc 3.4.6, _VERSION_ = "3.4.6 20060404 (Red Hat 3.4.6-3)"
4. ENVIRONMENT: x86_64/LP64 running linux-elf (Red Hat Enterprise Linux AS release 4 (Nahant Update 4)) with glibc 2.3
5. FILE: 22.locale.statics.mt.cpp
6. COMPILED: Aug 16 2007, 19:32:08
7. COMMENT: thread safety
   ############################################################

1. CLAUSE: lib.locale.statics

1. INFO (S1) (3 lines):
2. TEXT: testing std::locale::classic() with 2 threads
3. CLAUSE: lib.locale.statics

1. NOTE (S2) (4 lines):
2. TEXT: std::locale::global(const std::locale&) test disabled
3. CLAUSE: lib.locale.statics
4. LINE: 129

1. -----------------------------------------------+

2. DIAGNOSTIC

   ACTIVE

   TOTAL

   INACTIVE

3. -----------------------------------------------+

4. (S1) INFO

   2

   2

   0%

5. (S2) NOTE

   1

   2

   50%

6. -----------------------------------------------+

Application finished

_______________________________________________________________________________

ID	Short De	Seve	C	Context	Description	1st Acc	2nd Acce
	scriptio	rity	o	[Best]		ess[Bes	ss[Best]
	n	Name	u			t]
			n
			t

_______________________________________________________________________________

1	Write ->	Erro	3	"22.loc	Memory read at	"_mutex	"22.loca
	Read dat	r	6	ale.sta	"22.locale.statics.mt.cpp":78	.h":580	le.stati
	a-race			tics.mt	conflicts with a prior memory		cs.mt.cp
				.cpp":6	write at "_mutex.h":580 (flow		p":78
				9	dependence)

_______________________________________________________________________________

2	Read ->	Erro	1	[libstd	Memory write at [libstd15D.so,	"locale	[libstd1
	Write da	r		15D.so,	0xb646b] conflicts with a prior	_classi	5D.so,
	ta-race			0xb6460	memory read at	c.cpp":	0xb646b]
				]	"locale_classic.cpp":45 (anti	45
					dependence)

_______________________________________________________________________________

3	Read ->	Erro	1	[libstd	Memory write at [libstd15D.so,	"locale	[libstd1
	Write da	r		15D.so,	0xb646b] conflicts with a prior	_classi	5D.so,
	ta-race			0xb6460	memory read at	c.cpp":	0xb646b]
				]	"locale_classic.cpp":55 (anti	55
					dependence)

_______________________________________________________________________________

4	Thread t	Info	1	Whole	Thread termination at	"thread	"thread.
	erminati	rmat		Program	"thread.cpp":84 - includes stack	.cpp":8	cpp":84
	on	ion		1	allocation of 10.004 MB and use	4
					of 10.203 KB

_______________________________________________________________________________

5	Thread t	Info	1	Whole	Thread termination at	"thread	"thread.
	erminati	rmat		Program	"thread.cpp":84 - includes stack	.cpp":8	cpp":84
	on	ion		2	allocation of 10.004 MB and use	4
					of 10.188 KB

_______________________________________________________________________________

6	Thread t	Info	1	Whole	Thread termination at	"22.loc	"22.loca
	erminati	rmat		Program	"22.locale.statics.mt.cpp":171 -	ale.sta	le.stati
	on	ion		3	includes stack allocation of 10	tics.mt	cs.mt.cp
					MB and use of 8.094 KB	.cpp":1	p":171
						71

_______________________________________________________________________________

Comments

[Martin Sebor]

Attached Intel Thread Checker output in HTML format.

[Martin Sebor]

Attached a more informative Intel Thread Checker output file obtained with Intel C++ 10.0.

Attached Intel Thread Checker output after the committed change (http://svn.apache.org/viewvc?view=rev&rev=566892) and after __rw_once() has been locally doctored to use a global pthread mutex instead of pthread_once() to work around the Intel Thread Checker bug documented in STDCXX-467.

[Martin Sebor]

The absence of ITC data race errors depends on the linked bug in the tool (or a workaround for it).

[Martin Sebor]

In 4.2.0 we don't have time to work around the Thread Checker bug that causes it to issue the diagnostic. We might want to revisit it in 4.2.1. Rescheduled.

[Martin Sebor]

No time to deal with bugs in third party add-on tools. Deferred until 4.3.