[
https://issues.apache.org/jira/browse/AVRO-1033?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Douglas Creager updated AVRO-1033:
----------------------------------
Resolution: Fixed
Assignee: Douglas Creager
Status: Resolved (was: Patch Available)
Committed to SVN. I think a new JIRA ticket for the Item 2 is probably the
best way to go. Another option, besides bumping the required GCC version in
the preprocessor code, would be to perform an explicit autotools-style check in
the CMake build scripts to see whether the current compiler supports the GCC
intrinsics.
> Avro-C reference count fails on x86 and x86-64 for GCC 4.1.0 and earlier
> ------------------------------------------------------------------------
>
> Key: AVRO-1033
> URL: https://issues.apache.org/jira/browse/AVRO-1033
> Project: Avro
> Issue Type: Bug
> Components: c
> Affects Versions: 1.6.2
> Environment: Linux on an x86 or x86-64 platform with GCC version less
> than or equal to 4.1.0.
> Reporter: Vivek Nadkarni
> Assignee: Douglas Creager
> Fix For: 1.6.3
>
> Attachments: AVRO-1033.patch, test_refcount.c
>
> Original Estimate: 24h
> Remaining Estimate: 24h
>
> Item 1:
> *******
> In the Avro-C implementation, there is a bug in the Raw x86 assembly
> performing the atomic increment and decrement operations in
> refcount.h. Specifically, the assembly language commands (mistakenly)
> operate on the the pointer (refcount) instead of operating on the
> memory pointed to (*refcount). This results in memory leaks, when
> trying to deallocate avro values.
> This problem only occurs when using GCC versions less than 4.1.0. For
> GCC versions greater than 4.1.0, the code uses the GCC intrinsic
> __sync_add_and_fetch() or __sync_sub_and_fetch() instead of the raw
> x86 assembly, and this works properly.
> Since this problem is masked for developers working with new(er) GCC
> compilers, greater than version 4.1.0, one way to "uncover" it is to
> force the GCC intrinsics off by requiring a higher GCC version (say
> 8.1.0) to use the GCC intrinsics. Then GCC falls back to using the x86
> assembly, and the problem can be seen.
> I am attaching a sample test case (test_refcount.c), which will show a
> memory leak when compiled with the raw x86 commands enabled in
> refcount.h and run with valgrind using the following command:
> valgrind -v --leak-check=full --track-origins=yes ./test_refcount
> I am also attaching a simple patch to fix the bug. The patch just
> replaces (refcount) with (*refcount) in a few places, and the memory
> leak disappears.
> Item 2:
> *******
> On a related note, while compiling Avro across several flavors of
> Linux, we noticed that the GCC atomic intrinsics
> __sync_add_and_fetch() and __sync_sub_and_fetch() were not implemented
> for all versions of GCC greater than 4.1.0, causing compile errors.
> Specifically, the following two versions of GCC don't appear to
> support the GCC atomic intrinsics:
> 1. GCC 4.1.2 20080704 (Red Hat 4.1.2-48)
> 2. GCC 4.4.3 on Mandriva for an i586 target
> In our local code we have updated the GCC version test in refcount.h
> to require versions of GCC greater than 4.5.0, before including the
> GCC intrinsics. We might want to do the same in the global Avro
> repository. If appropriate, I can create a separate JIRA item for
> this issue. Please let me know.
> Cheers,
> Vivek Nadkarni
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
