Hi Leonid, Leonid Lisovskiy wrote,
On Thu, May 26, 2016 at 10:45 AM, Leonid Lisovskiy lly.dev@gmail.com wrote:
On Thu, May 26, 2016 at 12:41 AM, Waldemar Brodkorb wbx@uclibc-ng.org wrote:
Unfortunately this seems to break crisv32 support. (qemu-system-cris) The ex1 test is failing (never ends).
Any idea? On which arch you have tested?
I wasn't take into account that problem arise on platforms that has HAS_COMPARE_AND_SWAP defined only. cris is not in this list. Seems to problem come from original glibc sources and we have to fix __pthread_unlock() itself.
Will post new version of patch after some tests.
Thanks. Providing the CRIS image wouldn't be so simple, as neither strace nor gdbserver works at the moment. So I am not sure it would help.
Is there a reason you use LT on X86 or MIPS? Both have NPTL support. Normally LT is only enabled for noMMU architectures as BFIN/FR-V/C6X/Xtensa and for some exotic architectures AVR32/CRIS and for some architectures NPTL support wasn't ported from Glibc as M68k/Microblate/NIOS2.
Don't get me wrong, I am happy to have someone fixing LT bugs :) If you think LT is useful for NPTL supported architectures speak up.
best regards Waldemar