RFR (S): 8245509: Crash handler itself crashes when reporting Unsafe.putInt(0) crash

[David Holmes]

Hi Gerard,

On 29/05/2020 3:36 am, gerard ziemski wrote:
> hi all,
> 
> Please review this small and simple fix that adds ResourceMark to crash 
> handler when printing register info, without which can trigger fatal 
> error if the contents of a register happens to be an oop, which 
> uses as_C_string(), that in turn calls ResourceArea::allocate_bytes(), 
> which requires a ResourceMark.
> 
> If thread is not available, we do not attempt to print anything at this 
> step (i.e. print_register_info()), but the very next step (i.e. 
> print_context()) will always print the registers as HEX values.
> 
> bug link at https://bugs.openjdk.java.net/browse/JDK-8245509
> webrev at http://cr.openjdk.java.net/~gziemski/8245509_rev1
> passes Mach5 hs_tier1,2,3,4,5

This fixes the current problem and can be pushed.

But as Coleen notes in the bug report there are other callers of 
os::print_location that might also need a ResourceMark, and based on 
your stack trace it seems it is Universe::heap()->print_location that is 
the true source of the problem and where the ResourceMark perhaps should 
be? So I think we need a follow up bug to fix the broader problem.

Thanks,
David
-----

> 
> cheers