RFR(XS): 8079825: Uninitialised variable in hotspot/src/share/vm & cpu/x86/vm (runtime)

[Yumin Qi]

Calvin,

   Looks good.

Thanks
Yumin

On 7/16/2015 10:09 AM, Calvin Cheung wrote:
> This is for fixing some warnings about uninitialized variables 
> reported by our internal code analysis tool.
>
> JBS: https://bugs.openjdk.java.net/browse/JDK-8079825
>       (this is a confidential bug)
>
> webrev: http://cr.openjdk.java.net/~ccheung/8079825/webrev.00/
>
> [1] Uninitialised pointer localvariable_type_table_start may be used 
> as argument
> 8 when calling ClassFileParser::copy_localvariable_table at line 2465 of
> hotspot/src/share/vm/classfile/classFileParser.cpp in function
> ClassFileParser::parse_method
>
> Similar warning message for the other 3 local variables.
>
> The fix in classFileParser.cpp is for initializing the following 
> variables to NULL:
>     localvariable_table_length,
>     localvariable_table_start,
>     localvariable_type_table_length,
>     localvariable_type_table_start
>
> They may not be defined if the execution path goes through the if 
> block at line 2137:
>         if (LoadLineNumberTables &&
>             _cp->symbol_at(code_attribute_name_index) == 
> vmSymbols::tag_line_number_table()) {
>           // Parse and compress line number table
>           parse_linenumber_table(code_attribute_length, code_length,
>             &linenumber_table, CHECK_(nullHandle));
>
> [2] Uninitialised pointer rf may be used as argument 1 when calling
> RFrame::set_distance at line 96 of 
> hotspot/src/share/vm/runtime/rframe.cpp in
> function RFrame::new_RFrame
>
> The fix in rframe.cpp is for initializing the rf to NULL.
>
> It isn't being initialized if the execution path goes to the else block:
>     } else {
>       assert(false, "Unhandled frame type");
>     }
>
> the assert is only for debug build.
>
> Testing:
>    JPRT
>   Aurora adhoc run including vm.quick on linux, windows (32- and 
> 64-bit), and solaris sparc and x64
>
> Thanks in advance for your review.
>
> Calvin
>