Re: [PATCH roland/nptl] NPTL is no longer an add-on!

[Roland McGrath <roland at hack dot frob dot com>]

> i´ve checked out roland/nptl and build fails on s390 without appended 
> patch, because files are not found. So please add it to your branch.

Thanks.  I've included that with the latest rebase of the branch.

> make test|xtest runs without failures.

Great!

> -nptl/stamp.o: different ordering of object files

We don't care.

> -nptl/pthread_create.o:
> 	objdump -s -j .rodata.str1.2 nptl/pthread_create.o
> 	-after branch:
> ...
>   00e0 2d3e6865 61646572 2e746362 20213d20  ->header.tcb !=
>   00f0 2828766f 6964202a 29302900 2e2f6372  ((void *)0)../cr
>   0100 65617465 74687265 61642e63 00005f5f  eatethread.c..__
>   0110 7265636c 61696d5f 73746163 6b730000  reclaim_stacks..
>   0120 616c6c6f 63617465 5f737461 636b0000  allocate_stack..
>   0130 63726561 74655f74 68726561 64007374  create_thread.st
>   0140 6172745f 74687265 61640000           art_thread..
> 
> 	-before branch:
> ...
>   00e0 2d3e6865 61646572 2e746362 20213d20  ->header.tcb !=
>   00f0 2828766f 6964202a 29302900 2e2e2f6e  ((void *)0).../n
>   0100 70746c2f 63726561 74657468 72656164  ptl/createthread
>   0110 2e630000 5f5f7265 636c6169 6d5f7374  .c..__reclaim_st
>   0120 61636b73 0000616c 6c6f6361 74655f73  acks..allocate_s
>   0130 7461636b 00006372 65617465 5f746872  tack..create_thr
>   0140 65616400 73746172 745f7468 72656164  ead.start_thread
>   0150 0000                                 ..

Right.  Those are __assert_fail argument strings.  You can expect those to
change, and then the code may have some different offsets in some
instructions (either before linking or after or both) that refer to string
constants because the lengths changed.  Similarly, there may be
instructions setting up the integer argument for the line number and those
changing is also to be expected.

I've just done cross-builds for s390x-linux-gnu on master and (the newly
rebased and fixed) roland/nptl.  This is using a suspect cross-compiler
setup that has some bogons and fails to link a few things properly for some
reason, but does at least successfully compiles every object file.  In
those builds, I didn't see even those assert string differences.  But as I
say, my build is rather suspect and I'm nowhere near attempting to run
tests or anything (though ./elf/ld.so does start up and say hello under
qemu-s390x).

So please test again, but it sounds like we're now most likely all good for
s390.


Thanks,
Roland