[Bug 3341] [IcedTea8] jstack.stp should support ppc64[le,be]

[bugzilla-daemon at icedtea.classpath.org]

https://icedtea.classpath.org/bugzilla/show_bug.cgi?id=3341

--- Comment #3 from Mark Wielaard <mark at klomp.org> ---
(In reply to Gustavo Romero from comment #2)
> Patch looks good, I just have one question about it:
> 
> Why did you use value 72 in:
> 
>  fp = sp + 72; /* The on stack local variable storage.  */
> 
> Was that because you assumed  32 the size fo the "Parameter Save Area"?

That is a good question. The comment is wrong/misleading and the constant is
indeed magic. Sorry.

My first attempt got rid of fp, because ppc doesn't use a frame pointer and
given the stack pointer we could get directly at the parameter save area. But
that resulted in a lot of code changes just for ppc that had to be made
conditional, so I changed the approach to just have a "virtual frame pointer"
called fp, so the rest of the code could be used as is.

The actual unwinding doesn't need the frame pointer of course, we simply use
the stack and link pointers. And the pc is used to find the code block and
segment we want to inspect.

But the (virtual) frame pointer is used to get at the methodOop (for none
jit/native code blobs):

              // For compiled code the methodOop is part of the code blob.
              // For the interpreter (and other code blobs) it is on the
              // stack relative to the frame pointer.
              if (blob_name == "nmethod")
                methodPtr = @cast(blob, "nmethod",
                                     "@ABS_SERVER_LIBJVM_SO@")->_method
              else
                methodPtr = user_long(fp + (-3 * ptr_size)) & ptr_mask

So the above 72 really is 48 + (3 * 8) to create a "virtual frame pointer" that
acts as a real frame pointer on other architectures. Where 48 is the offset
from the stack pointer to the ppc64 parameter save area (the methodOop is the
first argument).

Hope that makes sense.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.java.net/pipermail/distro-pkg-dev/attachments/20190128/e0506977/attachment.html>