Fix pr61848, linux kernel miscompile

[Jeff Law]

On 09/16/14 00:03, Alan Modra wrote:
> gcc testsuite additions?  I decline.  It is too soon.  If you had read
> my patch submission you'll see that at some stage gcc was supposed to
> warn on conflicting section attributes, but hasn't done so for a very
> long time.  There needs to be some agreement on which direction we
> should go before I'm willing to spend even a small amount of time on
> the testsuite.
I think the direction for this ought to be quite clear, warn if we've 
got conflicting section attributes :-)  One could make the argument that 
if we had a test for conflicting section attributes that this code 
wouldn't have bit-rotted so badly.

I think for #1 in your original message, we should be issuing a warning 
of some kind as clearly something isn't right when we've asked for the 
same object to be in two different sections.




> Also, a test for merging tls model attributes runs
> into the problem that this can only be done in a target independent
> way by looking at dumps, and the tls model dump is currently broken.
Target independent tests are always best, but pulling together one that 
is specific to a target is usually still good enough.  Especially if 
it's an x86 target given how often those are tested by the developer 
community as a whole.

I think for #1 in your posting, verifying that we issue the proper 
warning should be sufficient.  Use the dg framework test and this marker:

/* { dg-require-named-sections } */

I'm not familiar enough with the TLS stuff to know what the right 
behaviour ought to be.

>
> Come to think of it, what if I decline to make any testsuite
> additions?  I'm asking because you're a steering committee member, and
> set policy.  I fully agree that testsuite additions are desirable,
> but, is *requiring* such testsuite additions a good idea?  Will that
> really improve gcc over time?  (Herding cats comes to mind.  I think
> you'll just drive away gcc contributors.)
There's no policy that requires testsuites, but for bugfixes most 
maintainers ask for them, or at least an explanation why they can't be 
provided.

GCC's testsuite is primarily a regression driven suite and these are 
great examples of behavior regressions that never should have happened, 
but did because of holes in the suite.  That's obviously something we 
ought to correct ;-)


Jeff