mysterious hang in pthread_create

[Kostik Belousov]

On Tue, Sep 02, 2008 at 09:38:20AM +0800, David Xu wrote:
> Kostik Belousov wrote:
> >On Mon, Sep 01, 2008 at 03:17:46PM +0300, Andriy Gapon wrote:
> >>on 01/09/2008 14:12 Kostik Belousov said the following:
> >>>On Mon, Sep 01, 2008 at 01:53:23PM +0300, Andriy Gapon wrote:
> >...
> >>>>>>>The following silly patch fixes the problem Andriy encountered:
> >>>>>>I can confirm - this fixed the issue for me.
> >>>>>>David, thanks!
> >>>>>Does libc_r still work with patch applied ?
> >>>>In what sense?
> >>>>The test program that I posted seems to hang in both cases (patched and 
> >>>>unpatched rtld).
> >>>The David' patch changes the code used to support libc_r operations.
> >>>Even on CURRENT, if you run 4.x-compiled binary with the support of
> >>>compat-4x libraries, this code from /libexec/ld-elf.so.1 (installed
> >>>by CURRENT buildworld) provides locking for rtld.
> >>I understand, but I am not sure what exactly needs to be tested.
> >>"Still works" is too broad in this context.
> >
> >I am not sure too. As I said, this one of the reason I prefered to not
> >touch that code.
> 
> It should not affect other code outside the rtld. In fact, this patch
> fixes the maintenance of synchronization between locks and thread_flag.
> In current code, if you acquire a rwlock, and then acquire second
> rwlock, the first one will work, but acquiring second lock will fail
> and thread_flag is out of sync, this results bit flag leaking in
> thread_flag, and later a _rtld_thread_init() call will transfer the
> unlocked rwlock state to libthr as locked state, also the existing
> code does not distinguish reader lock and writer lock, it blindly 
> transfer lock state as write-lock, fortunately, in correct case,
> it should not be called with any lock held, so the transferring
> does not occur.
Yes, I understand this. And, this code is used when threading implementation
is provided by libc_r, see above. I have no objection against that patch,
but I think that it shall be verified whether compat-4x threaded
programs work correctly with the change.

> Another question is why should dl_iterate_phdr() use exclusive lock ?
> doesn't this cause all C++ exception to be executed in serialization
> manner ?

It is required by C++ runtime. See the commit log for r178807.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-threads/attachments/20080902/1956f253/attachment.pgp