(jdk10) RFR(xxs): 8176872: [s390] wrong pc shown in error logs

Thomas Stüfe thomas.stuefe at gmail.com
Thu Mar 23 12:58:45 UTC 2017


a second webrev:

I forgot that the pc inside the secondary signal handler for crashes during
error reporting needs to be adjusted too.

Also, I need a second reviewer.

Thanks, Thomas

On Wed, Mar 22, 2017 at 9:05 AM, Thomas Stüfe <thomas.stuefe at gmail.com>

> Hi David, thanks for the review! Let's keep the s390-only solution for
> now. Cleanup can be done in a separate fix.
> Kind Regards, Thomas
> On Wed, Mar 22, 2017 at 12:47 AM, David Holmes <david.holmes at oracle.com>
> wrote:
>> Hi Thomas,
>> On 21/03/2017 11:40 PM, Thomas Stüfe wrote:
>>> Hi all,
>>> please take a look at this tiny fix. It fixes the pc shown as faulting
>>> address for SIGILL and SIGFPE in hs_err files.
>>> https://bugs.openjdk.java.net/browse/JDK-8176872
>>> http://cr.openjdk.java.net/~stuefe/webrevs/8176872-s390-wron
>>> g-pc-in-errorlogs/jdk10-webrev.00/webrev/
>>> When determining the crash pc, in all posix platform signal handlers pc
>>> is
>>> taken from the context. However, context.pc on zlinux points to the
>>> instruction *after* the faulting op. The correct way, according to POSIX,
>>> would be to take the address from siginfo_t->si_addr for signals SIGILL,
>> Well si_addr is the platform independent POSIX way of getting the
>> address. Getting it from the uc_mcontext is architecture specific and
>> seemingly not specified as to what the "pc" actually contains - the
>> faulting instruction or the next instruction. If it were specified then the
>> bug might be in zlinux :)
>> (actually, this patch would make sense for all POSIX platforms, but only
>>> s390 seems to show this error, so I leave the patch local to s390.)
>> I think it is historical that we get the PC from the uc_mcontext because
>> originally we did not have SA_SIGINFO available. So a broader cleanup of
>> this code is certainly possible. But for now this S390 fix seems fine.
>> Thanks,
>> David
>> -----
>> Kind Regards, Thomas

More information about the hotspot-runtime-dev mailing list