Re: [PATCH] parisc: futex: Use same lock set as lws calls

[John David Anglin <dave.anglin@bell.net>]

On 10/17/2011 11:47 AM, Carlos O'Donell wrote:
> The test-fork1 failure is still unexplained and happens intermittently.

I have built a lot of unstable on my rp3440.  I think this one causes 
failures in the thread
testsuites of perl, python2.7 and glib2.0.  These are characterized by 
tests hanging.

There is another class of failures.  They typically cause my rp3440 to 
crash due
to cache corruption.  The GCC libgomp and libatomic-ops testsuite seem 
to trigger
this one.  As I have mentioned, it's the libgomp "for" tests that
>
> The cancellation issues happen in tst-cancel*.
>
> I believe the cancellation issues are toolchain issues and I need to
> look into them.
Possibly, this is related to the following bug that I found last week 
building mpfr-3.1.0:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50691
A call to __tls_get_addr clobbers first argument of call to mpfr_cache.  
Don't have a
fix at the moment, but there is a simple testcase.

On a different subject, I tried to get udev-172-1 working.  However, 
this breaks bootstrap
due to an invalid argument in a call to inotify_init.  It's somewhat 
timing dependent since
some kernels will boot if they build enough of /dev before udev messes 
up.  In any case,
I believe that Guy Martin posted a patch a year or so ago to correct an 
inconsistency
between the glibc and the kernel for some bit definitions.  I'm thinking 
this may fix the
udev problem.

It appears that this change got lost and didn't make it into the debian 
glibc patch set.
Was a consensus reached on how to fix this inconsistency?

I plan on seeing if I can resolve the GCC mpfr bug, then I want to 
rebuild glibc with
the flag bits fixed.

I also found a GCC ICE compiling udev-172-1 with gcc-4.4.  Gcc-4.5 and 
gcc-4.6 seem
ok.

Dave

-- 
John David Anglin    dave.anglin@bell.net