threads/84483: problems with devel/nspr and -lc_r on 4.x

Mikhail T. mi at aldan.algebra.com
Tue Aug 2 14:00:41 GMT 2005


>Number:         84483
>Category:       threads
>Synopsis:       problems with devel/nspr and -lc_r on 4.x
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    freebsd-threads
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Tue Aug 02 14:00:37 GMT 2005
>Closed-Date:
>Last-Modified:
>Originator:     Mikhail T.
>Release:        FreeBSD 4.11-STABLE i386
>Organization:
Virtual Estates, Inc.
>Environment:
System: FreeBSD tulik.bychok.com 4.11-STABLE FreeBSD 4.11-STABLE #1: Wed Jul 20 21:04:19 EDT 2005 mi at tulik.bychok.com:/home/src/sys/compile/TULIK i386

>Description:

	The self-tests, which come with NSPR (devel/nspr port) are
	now patched and built to be meaningful.

	The also succeed on FreeBSD-5. On FreeBSD-4 some tests hang
	for ever all (or most) having to do with semaphores.

>How-To-Repeat:

	Update to the very latest version of the devel/nspr port.
	Run:
		make build test

	Some tests at the beginning may take a while, but the first
	one to truly hang is nameshm1:

	(gdb) where
	#0  0x280cfeac in semsys () from /usr/lib/libc_r.so.4
	#1  0x280c7122 in semop () from /usr/lib/libc_r.so.4
	#2  0x80526d8 in PR_WaitSemaphore ()
	#3  0x804c1d0 in ClientServerTest ()
	#4  0x804c643 in main ()
	#5  0x804b556 in _start ()

	You can kill the hung ones (killall nameshm1) and allow the
	test harness to proceed to the next one (randseed):

	(gdb) where
	#0  0x280dcf38 in __sys_poll () from /usr/lib/libc_r.so.4
	#1  0x280dc41d in _thread_kern_sched_state_unlock () from /usr/lib/libc_r.so.4
	#2  0x280dbdd2 in _thread_kern_scheduler () from /usr/lib/libc_r.so.4
	#3  0x0 in ?? ()

	A thread-developer needs to look at this to either fix -lc_r,
	or point out the bugs in the NSPR or its self-tests.

	The tests are single-file C-programs in

		work/nspr-4.6/mozilla/nsprpub/pr/tests

	they are run from their build-area in

		work/nspr-4.6/mozilla/nsprpub/build/pr/tests/

	Again, none of these problems exist on 5.x and, presumably,
	6.x (except for the instrumt-test, which crashes on all
	and is currently patched-out from the list).

>Fix:
>Release-Note:
>Audit-Trail:
>Unformatted:


More information about the freebsd-threads mailing list