[NPTL] Questions about testing

[Dan E]

On Wed, 22 Apr 2009 02:45 -0400, "Dan E" <trg_info at mailhaven.com> wrote:
> I dug deeper into this and it appears that the __clone function in
> libc/sysdeps/linux/mips/clone.S is missing a few parameters.  The first
> one that's missing is where the kernel should write the thread id.  See
> the comment about CLONE_PARENT_SETTID in
> libpthread/nptl/sysdeps/pthread/createthread.c.
> 
> I'm checking glibc's implementation and it looks like it handles all 7
> parameters.

OK, I fixed that up too based on the current glibc and the nptl tests
jumped from 52% to 94% passing.  169 out of 180 pass, with only 11
failing:

./tst-cancel16  FAIL
./tst-cancel7   FAIL
./tst-clock2    FAIL
./tst-cond20    FAIL
./tst-cond21    FAIL
./tst-flock1    FAIL
./tst-flock2    FAIL
./tst-mqueue1   FAIL
./tst-mqueue2   FAIL
./tst-mqueue4   FAIL
./tst-mqueue7   FAIL

It looks like NPTL on MIPS is getting a little closer to reality. 
libc/sysdeps/linux/mips/clone.S is definitely in the MIPS common code,
though, so perhaps you'd like to consider the following patch to see if
it breaks anything for the non-NPTL flavors of uClibc.

This patch modifies the MIPS userspace "clone()" function to accept the
three additional parameters that are now part of glibc.  Required for
NPTL pthread_create() to successfully obtain the thread id (tid) from
the kernel.


-- 
http://www.fastmail.fm - A no graphics, no pop-ups email service

-------------- next part --------------
A non-text attachment was scrubbed...
Name: mips-clone.patch
Type: text/x-patch
Size: 3712 bytes
Desc: not available
URL: <http://lists.busybox.net/pipermail/uclibc/attachments/20090422/743423be/attachment-0001.bin>