Correcting pointer access to top of stack

The current implementation always retrieves the deepest LR and TOC (belonging to very first patch) from the stack. Since, r5 contains base of stack and "decremented USED_SIZE" is not added to point to top of stack, so r5 keep pointing to bottom when retrieving LR/TOC values for any patched function.

lib/arch/powerpc64le/ulp_prologue.S

@@ -110,6 +110,10 @@ trampoline_routine:
   # Load ulp_stack ptr field.
   ld    %r5, ULP_STACK_PTR(%r5)
 
+  # Point to the top of stack but two, these two entries are popped in
+  # previous step and accessed in next step (stack size decremented before access).
+  add   %r5, %r5, %r6  # ulp_stack + used_size
+
   # Restore saved data.
   ld    %r2, 0(%r5)     # Restore TOC
   ld    %r8, 8(%r5)     # Restore LR