harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Evgueni Brevnov (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HARMONY-3649) [drlvm][jit] SPECjbb2005 intermittently fails on EM64T in server mode
Date Fri, 30 Nov 2007 12:42:43 GMT

     [ https://issues.apache.org/jira/browse/HARMONY-3649?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Evgueni Brevnov updated HARMONY-3649:
-------------------------------------

    Attachment: H3649.patch

As you can see from stack trace we have a problem with memory allocation by default STL allocator.
On Windows default STL allocator just delegates to system malloc/new. On Linux there is internal
STL implementation which may potentially have bugs, etc.  More over it's really strange to
have __simple_alloc::alloc on the stack since this is thread unsafe version of standart allocator.
To get rid of above problems I changed the code to use Jitrino internal allocator. Doing so
allows to be protected against multi-threading issues and STL bugs.

With that patch SpecJBB2005 didn't crashed over the night run. Build test checked on Lin ia32/em64t.

> [drlvm][jit] SPECjbb2005 intermittently fails on EM64T in server mode
> ---------------------------------------------------------------------
>
>                 Key: HARMONY-3649
>                 URL: https://issues.apache.org/jira/browse/HARMONY-3649
>             Project: Harmony
>          Issue Type: Bug
>          Components: DRLVM
>         Environment: EM64T
>            Reporter: George Timoshenko
>         Attachments: H3649.patch, stack
>
>
> The failure is very unstable.
> You can get only 1 failure from 10 runs.
> Or see 3 failures in a row.
> The failure allways looks like a segfault when some new object is being allocated in
STL code. (iterator, new item in a container, just a new std::string...)
> Usually the STL code is being called from execution manager or Jitrino.OPT code. But
the caller's code is OK and does not give any clue.
> the first time the failure was observed is April 4.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message