RFR (S) 7060111 : race condition in VMError::report_and_die()

[Volker Simonis]

Hi Frederic,

great to see this fixed! (I've submitted this bug and a patch for it
more than two years ago:)

I'd suggest to move the old comment for the fdSream up to the new
definition of  VMError::out like so:

+// An error could happen before tty is initialized or after it has been
+// destroyed. Here we use a very simple unbuffered fdStream for printing.
+// Only out.print_raw() and out.print_raw_cr() should be used, as other
+// printing methods need to allocate large buffer on stack. To format a
+// string, use jio_snprintf() with a static buffer or use staticBufferStream.
+fdStream VMError::out(defaultStream::output_fd());
+fdStream VMError::log; // error log used by VMError::report_and_die()

Otherwise the change looks good.

Thanks and best regards,
Volker

PS: would be great if you would not forget me in the 'Contributed-by' field :)

PPS: please use public bug-URLs like
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7060111 in the
future


On Thu, Jun 27, 2013 at 2:30 PM, frederic parain
<frederic.parain at oracle.com> wrote:
>
> Please review this small fix:
>
> http://cr.openjdk.java.net/~fparain/7060111/webrev.00/
>
> Bug: 7060111 : race condition in VMError::report_and_die()
>
> https://jbs.oracle.com/bugs/browse/JDK-7060111
>
> The bug is well described in its report. This fix uses
> class static fields initialized at library load time
> instead of static local variables lazily initialized.
>
> Test:
>
> The race condition is so tricky (2 threads must invoke
> VMError::report_and_die() almost at the same time), that I
> didn't try to produce a unit test for it. But the fix is
> so simple that it doesn't look risky.
>
> Thanks,
>
> Fred
>
> --
> Frederic Parain - Oracle
> Grenoble Engineering Center - France
> Phone: +33 4 76 18 81 17
> Email: Frederic.Parain at oracle.com