Re: Support six-argument syscalls from C for 32-bit x86, use generic lowlevellock-futex.h (bug 18138)

["Carlos O'Donell" <carlos at redhat dot com>]

On 03/23/2015 12:37 PM, Joseph Myers wrote:
> On Mon, 23 Mar 2015, Carlos O'Donell wrote:
> 
>> The last time I changed the 6-argument syscall code for hppa, I didn't 
>> notice I'd gotten it wrong until I reran more userspace tests like
>> running firefox with the new glibc.
>>
>> What kind of confidence do we have here without a test that exercises
>> the 6-argument syscall?
> 
> All the sem_timedwait tests (possibly tests for some other NPTL functions 
> as well) exercise this code through use of lll_futex_timed_wait_bitset.
> 
>> Is it possible to add an x86-specific test to show this is working
>> as intended?
> 
> The only six-argument syscall usage (inline from C, as opposed to from 
> syscalls.list) is in lowlevellock-futex.h, and I think the NPTL tests 
> cover that.  In fact only lll_futex_timed_wait_bitset gets used on x86 
> because of .S versions of pthread_cond_* where the other lll_futex_* 
> macros with six-argument syscalls get used in the C versions.
> 
> So I don't think there's any useful addition of tests possible here 
> (although of course if we do get future support for tests running in some 
> form of container where a test can safely change the system clock, then 
> adding the original test for bug 18138 to run in such a context makes 
> sense).

Thanks, I'm OK with this patch as-is then.

Unfortunately, with the thin container provisioning I'm thinking of using
for glibc testing, the system clock remains a shared resource from the host
that can't be changed.

We would need full qemu-based VM testing to test system clock changes.

Cheers,
Carlos.