RFR(xs): 8187028: [aix] Real thread stack size may be up to 64K smaller than requested one

[Lindenmaier, Goetz]

Hi,

offline:
was fixt denn der Change ueberhaupt? Also welche tests?
Und hast Du die hotspot / jdk jtreg tests laufen lassen? Gehen die durch?
Zumindest im Nachttest von jdk10/hs gibt’s gar kein Ergebnis auf aix.
Und auch wenn man nur die Tests anpassen muesste, eigentlich gehoert das
mit in Deinen change.

Gruesse,
  Goetz.



From: Thomas Stüfe [mailto:thomas.stuefe at gmail.com]
Sent: Thursday, August 31, 2017 7:44 PM
To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
Cc: ppc-aix-port-dev at openjdk.java.net
Subject: Re: RFR(xs): 8187028: [aix] Real thread stack size may be up to 64K smaller than requested one

Hi Goetz,

thank you for your review.

New Webrev:
http://cr.openjdk.java.net/~stuefe/webrevs/8187028-aix-Real-thread-stack-size-may-be-up-to-64K-smaller-than-requested-one/webrev.01/webrev/

Please find comments inline.

On Thu, Aug 31, 2017 at 3:54 PM, Lindenmaier, Goetz <goetz.lindenmaier at sap.com<mailto:goetz.lindenmaier at sap.com>> wrote:
Hi Thomas,

is it possible that AIX takes the os guard pages out of the stack in your case?
But if I remember correctly, AIX added them to the size of the stack correctly,
linux has non-standard behavior, see also create_thread in os_linux.cpp.


For several reasons I do not believe that is the case here. Among others, we set the guard page size at 4K so even if the OS deducts that from our stack size, it does not explain deltas of up to 64K.

Instead of thread->stack_base() - thread->stack_size(); please use stack_end().


Fixed.

Did you check this does not break the stack sizing tests? I think they start the
VM with small stacks and expect the VM not to start up. Now it might start
up if the bug is not on that platform because the stack is 64K bigger than
expected.


I changed to fix to only buff the thread stack size up for smaller thread stacks, to avoid running into overflow scenarios. As for too small stacks, I do not think it can be helped much. If these tests break, they should be adjusted, but they should not prevent this very reasonable fix.

Also, in the light of this bug, I do not think they can be made bullet proof. If the OS randomly hands out stack sizes which differ by as much as 64K, it is very difficult to precisely predict at which stack size stack overflows are going to happen and which size is still okay.

Please note that there is a wee bit of a time factor here, as jdk10/hs will close tomorrow for the next n weeks and I have vacations upcoming. I do not want to rush you. If you have objections, we will take time to address them, but the fix will have to wait until October if it is not pushed tomorrow.

Cheers, Thomas

Best regards,
  Goetz.




From: ppc-aix-port-dev [mailto:ppc-aix-port-dev-bounces at openjdk.java.net<mailto:ppc-aix-port-dev-bounces at openjdk.java.net>] On Behalf Of Thomas Stüfe
Sent: Thursday, August 31, 2017 12:09 PM
To: ppc-aix-port-dev at openjdk.java.net<mailto:ppc-aix-port-dev at openjdk.java.net>
Subject: RFR(xs): 8187028: [aix] Real thread stack size may be up to 64K smaller than requested one

Hi all,

please review this change:

Bug:
https://bugs.openjdk.java.net/browse/JDK-8187028

change:
http://cr.openjdk.java.net/~stuefe/webrevs/8187028-aix-Real-thread-stack-size-may-be-up-to-64K-smaller-than-requested-one/webrev.00/webrev/

The issue is that on AIX, pthread library seems to have a bug where it sometimes gives us less thread stack space than we requested (a variable amount, but seems to be 0..64K). This may cause intermittent stack overflow errors if the stacks are very small to begin with.

The workaround is to add 64K to the requested stack size to account for the fact that the OS may give us up to 64K less stack.

Also, improved logging.

Thanks, Thomas

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/ppc-aix-port-dev/attachments/20170831/3086c2b9/attachment-0001.html>