review (XS) for 7018101: os::dll_address_to_function_name returning wrong answers in 64 bit

Dmitry Samersoff Dmitry.Samersoff at oracle.com
Wed Feb 9 15:07:25 PST 2011


Tom,

Looks fine for me.

On 2011-02-10 00:39, Tom Rodriguez wrote:
> http://cr.openjdk.java.net/~never/7018101
>
> 7018101: os::dll_address_to_function_name returning wrong answers in 64 bit
> Reviewed-by:
>
> dladdr1 takes a pointer an Elf symbol structure through an untyped
> pointer.  The current code is using Sym which is typedef'ed to
> Elf32_Sym even in 64 bit mode.  This causes the result to be
> misinterpreted and produces bogus symbolic stack traces on 64 bit
> solaris.  It appears that the typedefs are done this way for
> compatibility reasons and that defining _ELF64 would cause the correct
> definitions to be used.  I can't find any documentation on that so
> instead I've chosen to explicitly pick the proper type.  Tested by
> producing hs_err files and comparing the symbolic trace the VM
> produces with what the external decoded produces.  This is for hs20.


-- 
Dmitry Samersoff
Java Hotspot development team, SPB04
* There will come soft rains ...


More information about the hotspot-dev mailing list