Generating oopmaps

[Tom Rodriguez]

Since you seem to emulating the interpreter I think you want to follow  
to same rules it uses.  If you are on a call bytecode then the portion  
of the expression stack containing arguments belongs to the callee.   
It's also possible to stop on a call site with the arguments set up  
and enter the runtime for call resolution and need to GC.  In that  
case there's special code in frame to handle it.  Look for the  
_arguments_do functions in frame.cpp since you might need to write  
your own.  And yes you have to record the oops in the monitor  
records.  If it's an oop and it's live you have to record it.

tom

On Jul 30, 2008, at 4:58 AM, Gary Benson wrote:

> Hi all,
>
> I'm writing a new compiler for HotSpot, and I'm trying to figure out
> what I need to include in my generated oopmaps.  My main issue is
> that, for all but the top frame, a frame may not know what is in some
> its own stack slots because they're now the locals of another frame.
> As an example, assuming my ABI stack looks something like this:
>
> ...
>
> 0xf7fbd640: stack[3]              = 0xd1ea8c00
> 0xf7fbd644: stack[2]              = 0xd1ea7738
> 0xf7fbd648: stack[1]              = 0xd1eaff38
> 0xf7fbd64c: stack[0]              = 0xd1eaff38
> 0xf7fbd650: method                = sun.tools.jar.Main.expand(Ljava/ 
> io/File;[Ljava/lang/String;Ljava/util/Vector;Ljava/util/Hashtable;)V
> 0xf7fbd654: frame_type            = SHARK_FRAME
> 0xf7fbd658: next_frame            = 0xf7fbd684
>
> 0xf7fbd65c: stack[7]              = 0xf7fbd61c = the top frame's  
> local[4]
> 0xf7fbd660: stack[6]              = 0xd1eafc98 = the top frame's  
> local[3]
> 0xf7fbd664: stack[5]              = 0x00000002 = the top frame's  
> local[2]
> 0xf7fbd668: stack[4]              = 0xd1ec0450 = the top frame's  
> local[1]
> 0xf7fbd66c: stack[3]              = 0xd1ec0438 = the top frame's  
> local[0]
> 0xf7fbd670: stack[2]              = 0xd1eabf38
> 0xf7fbd674: stack[1]              = 0xd1ea7738
> 0xf7fbd678: stack[0]              = 0xd1ec02d8
> 0xf7fbd67c: method                = sun.tools.jar.Main.expand(Ljava/ 
> io/File;[Ljava/lang/String;Ljava/util/Vector;Ljava/util/Hashtable;)V
> 0xf7fbd680: frame_type            = SHARK_FRAME
> 0xf7fbd684: next_frame            = 0xf7fbd6b0
>
> 0xf7fbd688: stack[7]              = 0xd1ea7b50 = the middle frame's  
> local[4]
> 0xf7fbd68c: stack[6]              = 0xd1ea7738 = the middle frame's  
> local[3]
> 0xf7fbd690: stack[5]              = 0x00000002 = the middle frame's  
> local[2]
> 0xf7fbd694: stack[4]              = 0xd1ec0450 = the middle frame's  
> local[1]
> 0xf7fbd698: stack[3]              = 0xd1ec0438 = the middle frame's  
> local[0]
> 0xf7fbd69c: stack[2]              = 0xd1ea7728
> 0xf7fbd6a0: stack[1]              = 0xd1ea70c0
> 0xf7fbd6a4: stack[0]              = 0xd1ec02d8
> 0xf7fbd6a8: method                = sun.tools.jar.Main.expand(Ljava/ 
> io/File;[Ljava/lang/String;Ljava/util/Vector;Ljava/util/Hashtable;)V
> 0xf7fbd6ac: frame_type            = SHARK_FRAME
> 0xf7fbd6b0: next_frame            = 0xf7fbd720
>
> ...
>
> Would it be correct for the oopmap of the middle frame (ie
> 0xf7fbd65c-0xf7fbd684) to include oops in the ranges
> 0xf7fbd670-0xf7fbd678 and 0xf7fbd688-0xf7fbd698) but to not
> include oops in the range 0xf7fbd65c-0xf7fbd66c?
>
> Another question is, when there are monitors in the frames,
> should I include the objects in the monitors in the oopmaps
> I generate?
>
> Thanks in advance!
>
> Cheers,
> Gary
>
> -- 
> http://gbenson.net/