libthr and main thread stack size

[John Baldwin]

On Tuesday, September 16, 2014 11:13:24 AM Konstantin Belousov wrote:
> On Mon, Sep 15, 2014 at 03:47:41PM -0600, Justin T. Gibbs wrote:
> > On Aug 8, 2014, at 5:22 AM, Konstantin Belousov <kostikbel at gmail.com>
> > wrote:
> > 
> > ?
> > 
> > > Below is the patch which adds environment variable
> > > LIBPTHREAD_BIGSTACK_MAIN. Setting it to any value results in the
> > > main thread stack left as is, and other threads allocate stack
> > > below the area of RLIMIT_STACK. Try it. I do not want to set this
> > > behaviour as default.
> > 
> > Is there a reason this should not be the default? Looking at the
> > getrlimit() page on the OpenGroup?s site they say:
> > 
> > RLIMIT_STACK This is the maximum size of the initial thread's stack,
> > in bytes. The implementation does not automatically grow the stack
> > beyond this limit. If this limit is exceeded, SIGSEGV shall be
> > generated for the thread. If the thread is blocking SIGSEGV, or the
> > process is ignoring or catching SIGSEGV and has not made arrangements
> > to use an alternate stack, the disposition of SIGSEGV shall be set to
> > SIG_DFL before it is generated.
> > 
> > Does posix say something different?
> > 
> > I ran into this issue when debugging a segfault on Postgres when
> > running an (arguably quite bogus) query that should have fit within
> > both the configured stack rlimit and Postgres? configured stack limit.
> > The Postgres backend is really just single threaded, but happens
> > to pull in libpthread due to the threading support in some of the
> > libraries it uses. The segfault definitely violates POLA.
> > 
> > ? Justin
> 
> I am conservative to not disturb the address space layout in single go.
> If enough people test this setting, I can consider flipping the default
> to the reverse.
> 
> I am still curious why the things were done in this way, but nobody
> replied.

I suspect it was done out of reasons of being overly conservative in 
interpreting RLIMIT_STACK.  I think it is quite surprising behavior though and 
would rather we make your option the default and implement what the Open Group 
says above.

-- 
John Baldwin