dl_iterate_phdr() deadlock patch

[Boehm, Hans]

A quick status update:

I checked the patch into the trunk.

I then decided to run SPECjbb on X86 as one last test before checking it into 3.3.  That failed.  It turns out that dl_iterate_phdr() apparently doesn't issue a callback for the main executable in a completely statically linked program, though it appears to do so if the program is linked against any dynamic libraries.

I am generating a small patch to work around this.  Unless I hear otherwise, I will check the result into 3.3 very soon, and update the trunk correspondingly.

Hans

> -----Original Message-----
> From: Tom Tromey [mailto:tromey@redhat.com]
> Sent: Saturday, March 01, 2003 3:48 PM
> To: Boehm, Hans
> Cc: 'java-patches@gcc.gnu.org'
> Subject: Re: dl_iterate_phdr() deadlock patch
> 
> 
> >>>>> "Hans" == Boehm, Hans <hans_boehm@hp.com> writes:
> 
> Hans> This patch is extracted from my current GC tree.  I would like
> Hans> to check it into both the trunk and 3.3.  It should do 4 things:
> 
> Thanks.  I'm still hoping to import the new GC, but it seems less and
> less likely before 3.3 due to lack of time.
> 
> This patch was the major thing we wanted from the import.  So let's
> put it in to 3.3 and 3.4.  It won't hurt if I'm able to do the import,
> and it will help a lot of I'm not able.
> 
> Hans> Question: Is it currently true that GC_dlopen can only be called
> Hans> from one thread at a time?
> 
> Yes, we synchronize around ltdl calls.
> 
> Well, SharedLibLoader isn't synchronized, but that isn't used much and
> has some bugs anyway.  Though this reminds me I need to fix up my big
> solib: patch to make sure the dlopen is synchronized properly.  I put
> this on my to-do list.
> 
> Tom
>