Hi, On 6/3/19 11:36 AM, Nick Gasson wrote: > >> We need to know what it's used for to know which solution is right. I'm >> guessing the primary use case is asynchronous runtime stack probes, used >> by debugging tools. >> > > Yes, we also have os::stack_shadow_pages_available() which uses it to > calculate how much space there is between the current SP and the end of > the stack. In this case I think it's ok as long as we don't return a > value that's *above* the actual stack pointer. And os::current_frame(), > which constructs a `frame' object using the FP of the caller, but the SP > will point into the frame of os::current_frame(), so it seems it's > already inaccurate.
Eww. > There's also a comment in os_linux.cpp that says "Don't use > os::current_stack_pointer(), as its result can be slightly below current > stack pointer". So I'm wondering if we can simplify this a lot and use > __builtin_frame_address(0) which will give us the FP in > os::current_stack_pointer (ought to be the caller's SP - 16). Zero does > this currently. And maybe we can replace _get_previous_fp() with > __builtin_frame_address(1)? Let's make os::current_stack_pointer() noinline, then make it return __builtin_frame_address(0). -- Andrew Haley Java Platform Lead Engineer Red Hat UK Ltd. <https://www.redhat.com> https://keybase.io/andrewhaley EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671