RFR: 8294160: misc crash dump improvements

[David Holmes]

On Thu, 22 Sep 2022 07:53:42 GMT, Dean Long <dlong at openjdk.org> wrote:

> Try to recover and output meaningful information in more situations, such as:
> - when a pointer is in Method metadata memory range, but the memory is not committed
> - when we call a bad address
> 
> When calling a bad address, we usually don't get a meaningful stack backtrace, but in many situations we can if we know where to find the caller information.

I'm not familiar with all the aspects of this in enough detail so just a couple of observational comments.

src/hotspot/os/posix/signals_posix.cpp line 629:

> 627:   // Check for UD trap caused by NOP patching.
> 628:   // If it is, patch return address to be deopt handler.
> 629:   if (!signal_was_handled && pc != NULL && os::is_readable_pointer(pc)) {

If these new conditions are not met then what happens? What code will get the unreadable, or null pc to process?

src/hotspot/share/utilities/vmError.cpp line 353:

> 351: 
> 352:   // see if it's a valid frame
> 353:   if (fr.sp() != nullptr && os::is_readable_pointer(fr.sp())) {

Why did you completely change the notion of "valid frame" here?

-------------

PR: https://git.openjdk.org/jdk/pull/10387