valgrind on 5.3BETA1

[Dan Nelson]

In the last episode (Aug 23), Simon Barner said:
> Miguel Mendez wrote:
> > Is anybody using valgrind on 5.3BETA1 or newer? I've been using it
> > for a while on 5.2.1 without any problems, until I upgraded my
> > system to RELENG_5. The program works but it never exits after
> > running your program, it gets stuck into `umtx' state indefinitely.
> > This is happening with both plain jane valgrind and
> > valgrind-snapshot ports. Is this already a known issue?
> 
> Hi,
> 
> there is a closed problem report in the database:
> http://www.freebsd.org/cgi/query-pr.cgi?pr=68992
> 
> At the end of that PR, there is a patch that seems to work around the
> problem. Could you please test it?

That patch looks like it's already in the valgrind-snapshot-352 port,
and it still hangs on me.  Adding some debugging to the code that calls
the umtx syscalls give me:

$ valgrind --skin=memcheck date
==14620== Memcheck, a memory error detector for x86-linux.
==14620== Copyright (C) 2002-2004, and GNU GPL'd, by Julian Seward.
==14620== Using valgrind-2.1.2.CVS, a program supervision framework for x86-linux.
==14620== Copyright (C) 2000-2004, and GNU GPL'd, by Julian Seward.
==14620== Locking#1 mutex 0x0/0x0
==14620== Done 0x187B8/0x187B8
==14620== For more details, rerun with: -v
==14620==
Mon Aug 23 16:17:01 CDT 2004
==14620==
==14620== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
==14620== malloc/free: in use at exit: 4096 bytes in 1 blocks.
==14620== malloc/free: 1 allocs, 0 frees, 4096 bytes allocated.
==14620== For a detailed leak analysis,  rerun with: --leak-check=yes
==14620== For counts of detected errors, rerun with: -v
==14620== Unlocking mutex 0x187B9/0x187B9
==14620== Done 0x187B9/0x187B9
==14620== Locking#2 mutex 0x187B9/0x187B9
( send SIGKILL from another vty here )
zsh: 14620 killed     valgrind --skin=memcheck date

There doesn't seem to be any docs on these umtx syscalls so I don't
know what the expected behaviour is supposed to be.  I printed the
pointer and the contents of the u_owner member before and after the
syscall.

-- 
	Dan Nelson
	dnelson at allantgroup.com
-------------- next part --------------
--- coregrind/vg_proxylwp.c~	Mon Aug 23 15:47:33 2004
+++ coregrind/vg_proxylwp.c	Mon Aug 23 15:48:42 2004
@@ -596,6 +596,8 @@
    static const vki_kstack_t ss = { .ss_flags = VKI_SS_DISABLE };
 
 #if __FreeBSD__ == 5
+   VG_(message)(Vg_UserMsg, "Locking#1 mutex %p/%p", px->mutex, px->mutex.u_owner);
    VG_(do_syscall)(__NR__umtx_lock, &px->mutex);
+   VG_(message)(Vg_UserMsg, "Done %p/%p", px->mutex, px->mutex.u_owner);
    px->mutex.u_owner |= UMTX_CONTESTED;
    VG_(do_syscall)(__NR_thr_self, &px->lwp);
@@ -943,6 +944,8 @@
    px->exitcode = ret;
 
 #if __FreeBSD__ == 5
+   VG_(message)(Vg_UserMsg, "Unlocking mutex %p/%p", px->mutex, px->mutex.u_owner);
    ret = VG_(do_syscall)(__NR__umtx_unlock, &px->mutex);
+   VG_(message)(Vg_UserMsg, "Done %p/%p", px->mutex, px->mutex.u_owner);
    if (use_rfork)
       VG_(do_syscall)(__NR_exit, 0);
@@ -1171,7 +1173,11 @@
 #if __FreeBSD__ == 5
    if (block) {
        if(proxy->lwp != 0)
+       {
+	   VG_(message)(Vg_UserMsg, "Locking#2 mutex %p/%p", proxy->mutex, proxy->mutex.u_owner);
 	   VG_(do_syscall)(__NR__umtx_lock, &proxy->mutex);
+	   VG_(message)(Vg_UserMsg, "Done %p/%p", proxy->mutex, proxy->mutex.u_owner);
+       }
 
        if (status)
 	   *status = proxy->exitcode;