RFR: 8199323: hsdis could not be loaded which are located on long path (original) (raw)

David Holmes david.holmes at oracle.com
Tue Mar 13 11:22:17 UTC 2018

On 13/03/2018 6:26 PM, Yasumasa Suenaga wrote:

Thanks David!

I've run test on submit-hs repo, but I received 1 failure: mach5-one-ysuenaga-JDK-8199323-20180313-0429-14193 java/lang/invoke/condy/CondyInterfaceWithOverpassMethods.java windows-x64 Error: failed to clean up files after test I guess the failure does not relate to this change.

No. "failed to clean up files after test" is something we tend to see a bit on windows.

After getting second reviewer, I re-run the test before pushing.

Ok.

David

Yasumasa

2018-03-13 16:38 GMT+09:00 David Holmes <david.holmes at oracle.com>: Looks fine to me. Just need a second review. And if you use the new submit-hs repo [1] to do pre-push testing you can push this yourself. Thanks, David [1] http://mail.openjdk.java.net/pipermail/hotspot-dev/2018-March/030656.html

On 13/03/2018 12:25 AM, Yasumasa Suenaga wrote: Hi David,

I don't think this code has the same concern that the code in jvmmd.h claims** to have, so a simple use of MAXPATHLEN should be fine on all non-windows platforms.

It sounds good to me. I updated webrev: http://cr.openjdk.java.net/~ysuenaga/JDK-8199323/webrev.01/

My only concern with the current change is whether a 4K on stack buffer might cause any issues? In case of HotSpot for x64 Linux, stack size is 1MB. So I think the risk of stack overflow is very low. In fact, my environment (Fedora 27 x64) works fine with this change. Thanks, Yasumasa On 2018/03/12 13:13, David Holmes wrote: Hi Yasumasa, On 8/03/2018 11:21 PM, Yasumasa Suenaga wrote: Hi all, Could you review and sponsor it? webrev: http://cr.openjdk.java.net/~ysuenaga/JDK-8199323/webrev.00/ JBS: https://bugs.openjdk.java.net/browse/JDK-8199323 Mach5 test result on submit repo: mach5-one-ysuenaga-JDK-8199323-20180308-1027-13701 I encountered DebuggerException when hsdis is located on long path as below: Location of hsdis: /home/yasuenag/work/xxxxxx/xxxxxxxxxxxxxx/xxxxxxxxxxxxx/workspace/usr/lib/jvm/java-1.8.0-openjdk-1.8.0.101-3.b13.el68.x8664/jre/lib/amd64/hsdis-amd64.so Exception: sun.jvm.hotspot.debugger.DebuggerException: /home/yasuenag/work/xxxxxx/xxxxxxxxxxxxxx/xxxxxxxxxxxxx/workspace/usr/lib/jvm/java-1.8.0-openjdk-1.8.0.101-3.b13.el68.x8664/j: cannot open shared object file: No such file or directory In JavasunjvmhotspotasmDisassemblerload1library(), buffer which uses for library path is defined as below: _ _char buffer[128];_ _ I copied JVMMAXPATHLEN related code to sadis.c from os/posix/include/jvmmd.h and os/windows/include/jvmmd.h .

I don't think this code has the same concern that the code in jvmmd.h claims** to have, so a simple use of MAXPATHLEN should be fine on all non-windows platforms. ** The posix jvmmd.h code is historical and I don't think we have to be concerned either about a 4095 definition of MAXPATHLEN or that the VM and libraries may have been compiled on different Linux versions! My only concern with the current change is whether a 4K on stack buffer might cause any issues? Thanks, David -----

I added noreg-hard label on this ticket because this issue is available when disassembling on coredump.

Thanks, Yasumasa

More information about the serviceability-dev mailing list