On Tue, Aug 30, 2005 at 10:59:25AM -0700, prasad wrote:
> I am posting the email message from Jonathan to the forum:
> (One of these days I'll subscribe to the list)
>
> -------------------------------------------------------------------------------------------------------------
> On Tue, Aug 30, 2005 at 06:34:41AM -0700, prasad jlv wrote:
> >
> >
> > Here's the output from ::umem_status
> > -------------------------------------------------------
> > > ::umem_status
> > Status: ready and active
> > Concurrency: 8
> > Logs: (inactive)
> > Message buffer:
> > recursive allocation during getenv(3C) calls --
> > getenv(3C) results ignored.
>
> Well, that would definitely be the problem, then; here's the
> troublesome path:
>
> > /1 at 1: -> libumem:umem_setup_envvars(0x0, 0x0,
> > 0xfece8bc0, 0xff3a2000)
> > /1 at 1: -> libc:getenv(0xff37742c, 0x0, 0x0, 0x0)
> > /1 at 1: <- libc:getenv() = 0xffbffb04
> > /1 at 1: -> libc:getenv(0xff37744c, 0x0, 0x9fc4c,
> > 0x0)
> > /1 at 1: <- libc:getenv() = 0
> > /1 at 1: -> libc:getenv(0xff377470, 0x1, 0x9fc4c,
> > 0x0)
> > /1 at 1: <- libc:getenv() = 0
> **
> > /1 at 1: -> libumem:umem_setup_envvars(0x1, 0x0,
> > 0xfece8bc0, 0xff3a2000)
> > /1 at 1: <- libumem:umem_setup_envvars() = 1
> > /1 at 1: <- libumem:umem_setup_envvars() = 0xff211b40
>
> ** is where we recurse; looking at the code, we're recursing in:
>
> if ((h = dlopen(0, RTLD_FIRST | RTLD_LAZY)) != NULL) {
>
> there's a bug in the recursion detection code; it assumes that the
> dlopen
> call can't trigger '.init' sections. I'll investigate and get back to
> you.
This is a side-effect of the following dynamic linker bug:
6301218 Matlab dumps core on startup when running on 112963-22
You need to pick up the latest dynamic linker patch; I'm not sure of the
patch-id. (112963 is for Solaris 9)
Cheers,
- jonathan
--
Jonathan Adams, Solaris Kernel Development
