[Bug libc/11787] Program with large TLS segment fails aio_write

["carlos_odonell at mentor dot com" <sourceware-bugzilla at sourceware dot org>]

http://sourceware.org/bugzilla/show_bug.cgi?id=11787

--- Comment #4 from Carlos O'Donell <carlos_odonell at mentor dot com> 2012-03-23 20:34:37 UTC ---
I've reviewed the patch sent to the mailing list and provided comments:
http://sourceware.org/ml/libc-alpha/2012-03/msg01002.html

I just noticed that what I recommended shouldn't be required since the code
should already take this into account.

Nobody has given any hard numbers about the static TLS size so I'm marking this
issue as unconfirmed.

Please provide some real-world figures so we know how bad the problem is
compared to the default stack size for x86 e.g. 2MB.
sysdeps/i386/pthreaddef.h:#define ARCH_STACK_DEFAULT_SIZE       (2 * 1024 *
1024)

For example code in nptl-init.c tries to take things into account:
~~~ nptl/nptl-inic.c:
414   /* Determine the default allowed stack size.  This is the size used
415      in case the user does not specify one.  */
416   struct rlimit limit;
417   if (getrlimit (RLIMIT_STACK, &limit) != 0
418       || limit.rlim_cur == RLIM_INFINITY)
419     /* The system limit is not usable.  Use an architecture-specific
420        default.  */
421     limit.rlim_cur = ARCH_STACK_DEFAULT_SIZE;
422   else if (limit.rlim_cur < PTHREAD_STACK_MIN)
423     /* The system limit is unusably small.
424        Use the minimal size acceptable.  */
425     limit.rlim_cur = PTHREAD_STACK_MIN;
426
427   /* Make sure it meets the minimum size that allocate_stack
428      (allocatestack.c) will demand, which depends on the page size.  */
429   const uintptr_t pagesz = GLRO(dl_pagesize);
430   const size_t minstack = pagesz + __static_tls_size + MINIMAL_REST_STACK;
431   if (limit.rlim_cur < minstack)
432     limit.rlim_cur = minstack;
433
434   /* Round the resource limit up to page size.  */
435   limit.rlim_cur = (limit.rlim_cur + pagesz - 1) & -pagesz;
436   __default_stacksize = limit.rlim_cur;
~~~

Note that we check to see that __default_stacksize takes into account
__static_tls_size and a minimal rest size (2K on x86).

Is the problem that the minimal computed is *still* not enough?

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.