[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-hotspot-runtime-dev
Subject:    VMError and masked signals
From:       coleen.phillimore () oracle ! com (Coleen Phillimore)
Date:       2010-11-10 15:06:26
Message-ID: 4CDAB4F2.9060304 () oracle ! com
[Download RAW message or body]

On 11/9/2010 11:06 PM, Tom Rodriguez wrote:
> I was looking at a crash Vladimir saw with a truncated hs_err file when debugging a \
> SIGFPE.  What happens is that during the reporting in VMError is tries to print the \
> native stack and gets a SEGV.  Instead of restarting the reporting like it normally \
> does the process just aborts.  I looked into it a bit and it appears that it's \
> because we don't unmask SEGV and BUS when we're running the VMError stuff so if we \
> get one of them while in the signal handler, the OS simply kills the process.  \
> Normally all signals are masked except the one being handled so we will handle a \
> SEGV when reporting a SEGV or BUS when reporting a BUS but we'll be killed the OS \
> when they are mixed.  I think VMError::reset_signal_handler and crash_handler \
> should be unblocking SEGV and BUS during reporting if we want to get complete \
> hs_err files.  Does anyone see any problems with this? 
> tom
Not at all.

Coleen


[prev in list] [next in list] [prev in thread] [next in thread]