RFC: 8216472: Stack overflow followed by crash

[Alex Kashchenko]

On 11/08/2019 11:05 AM, Alan Bateman wrote:
> On 04/11/2019 22:29, Alex Kashchenko wrote:
>>
>> Renamed and simplified the test. In this form it does not produce 
>> hs_err files anymore (stack overflow happens inside mswsock.dll and OS 
>> kills the process). Changes to WindowsSelectorImpl are left the same:
>>
>> https://cr.openjdk.java.net/~akasko/jdk/8216472/webrev.02/
>>
> Thanks for the test rename and cleanup but are you saying that it 
> doesn't reproduce with this test?

Crash itself is reproduced, I am getting it reliably (and it happens 
much faster with simplified test). Just the hs_err file is not written 
by JVM. When being run directly, system dialog "Program Has Stopped 
Working" is displayed instead with details pointing to mswsock.dll (that 
is called in select() call). For jtreg runs this new behaviour seems not 
to not change much (process dies and test fails with both old and new 
variants).

With initial test, hs_err was written most of the times, but not always. 
As I understand it is not written if process steps over the "red stack 
zone" and is killed by OS immediately.

-- 
-Alex