Spontaneuous crashes with OpenJDK6
Tom Rodriguez
tom.rodriguez at oracle.com
Tue May 17 10:38:28 PDT 2011
Those are odd looking crashes. Decoding the assembly shows this:
siginfo:si_signo=SIGSEGV: si_errno=0, si_code=128 (), si_addr=0x00000000;;
Registers:
EAX=0x00000001, EBX=0xbfda42f0, ECX=0x930d59d0, EDX=0x917bb158
ESP=0xbfda42f0, EBP=0xbfda43cc, ESI=0xb5febcb0, EDI=0x00000001
EIP=0xb5febcd3, CR2=0xb7433100, EFLAGS=0x00210202
;; b5febcc3 8b 54 24 38 mov 0x38(%esp,1),%edx
;; b5febcc7 8b 4c 24 54 mov 0x54(%esp,1),%ecx
;; b5febccb 0f 84 15 00 00 00 je 0xb5febce6
;; b5febcd1 90 nop
;; b5febcd2 90 nop
;; ---------------
;; b5febcd3 e8 f8 63 52 00 call 0xb65120d0
;; b5febcd8 83 f8 00 cmp $0x0,%eax
;; b5febcdb 0f 84 1d 00 00 00 je 0xb5febcfe
;; b5febce1 e9 33 ff ff ff jmp 0xb5febc19
How does a call generate a SEGV? si_code=128 seems to be SI_KERNEL which I guess means it was sent from an interactive terminal? This looks like some sort of OS issue to me. I've never seen anything like it.
Many of the logs don't have the instructions because of bugs in the error dumping in hs19 but here's another one:
;; b5457d46 00
;; b5457d47 00 00 add %al,(%eax)
;; b5457d49 0f be 52 08 movsbl 0x8(%edx),%edx
;; b5457d4d 83 fa 00 cmp $0x0,%edx
;; b5457d50 0f 84 2e 03 00 00 je 0xb5458084
;; ---------------
;; b5457d56 e9 19 03 00 00 jmp 0xb5458074
;; b5457d5b 8b 54 24 38 mov 0x38(%esp,1),%edx
;; b5457d5f 8b 4c 24 3c mov 0x3c(%esp,1),%ecx
;; b5457d63 90 nop
;; b5457d64 90 nop
;; b5457d65 90 nop
Very odd. A little googling for "si_code=128" SIGSEGV suggests that this problem has been around for a while on Linux. Some of the crashes are even in C++ code both inside and outside of the JVM. Any Linux gurus know what a signal like this means?
tom
On May 17, 2011, at 2:18 AM, Clemens Eisserer wrote:
> Hi,
>
> For some time now I experience sporadic crashes running eclipse and
> netbeans on my i686 system with the client compiler.
> Contrary to what is stated in the bug-report, it crashes also when
> using the serial GC.
> According to hotspot's crash log, the crash always happens while
> executing compiled java-code.
>
> I know OpenJDK is unsupported, but in case somebody is interested:
> https://bugzilla.redhat.com/show_bug.cgi?id=677730
>
> Thanks, Clemens
More information about the hotspot-runtime-dev
mailing list