Menu
Our JVM crashes with segmentation fault from time to time in production with what feels like a race condition of some sort.
Setups to reproduce: - JRE jre1.6.0_24 on Linux Ubuntu 9.10 and Debian 4.x 64 bit multicore AMD - Apache tomcat 6.0.24 ... 6.0.32
Recompiling java with fastdebug reproduces the problem (this gcc -g1). However, it doesn't yield much more useful information then what we have here.
Recompiling java with debug does not reproduce the problem (this is gcc -g plus possibly some -DSOMETHING code debug flags).
Any help trying to figure this out would be most appreciated.
Core file generated using the 1.6.0_24 jdk from Oracle . gdb turns up:
Program terminated with signal 11, Segmentation fault.
#0 0x00002aaab7b10666 in NET_Read () from
/usr/local/jdk1.6.0_24/jre/lib/amd64/libnet.so
Ok, so my assembly is really really really rusty. Keeping that in mind:
(gdb) info frame
Stack level 0, frame at 0x4b3e0040:
rip = 0x2aaab7b10666 in NET_Read; saved rip 0x2aaab7b0d53b
called by frame at 0x4b3f0090
Arglist at 0x4b3dffc8, args:
Locals at 0x4b3dffc8, Previous frame's sp is 0x4b3e0040
Saved registers:
rbx at 0x4b3e0008, rbp at 0x4b3e0010, r12 at 0x4b3e0018, r13 at
0x4b3e0020, r14 at 0x4b3e0028, r15 at 0x4b3e0030, rip at 0x4b3e0038
So, gdb tells us that the argument list is at:
0x4b3dffc8
Looking at the data there:
(gdb) x/8x 0x4b3dffc8
0x4b3dffc8: 0x00000000 0x00000000 0x00000000 0x00000000
0x4b3dffd8: 0x00000000 0x00000000 0x00000000 0x00000000
`
So no dice there. Again, my assembly dates back to 2nd wave ska, so I can only think that either the stack is somewhat buggered, or the gcc optimization flags generate code which uses registers for arguments instead of the stack?
Onto the registers:
(gdb) info registers
rax 0xf2 242
rbx 0x4 4
rcx 0x2b73aa8bfed3 47775782534867
rdx 0x4 4
rsi 0x4b3e0050 1262354512
rdi 0xf2 242
rbp 0x0 0x0
rsp 0x4b3dffd0 0x4b3dffd0
r8 0xffc 4092
r9 0x2b73aa8c61b0 47775782560176
r10 0x2b73aa8c9f78 47775782575992
r11 0x2b73aa8b20d0 47775782478032
r12 0xf2 242
r13 0xf2 242
r14 0x2aaabad4b9c8 46912767310280
r15 0x4 4
rip 0x2aaab7b10666 0x2aaab7b10666 <NET_Read+22>
eflags 0x10202 [ IF RF ]
cs 0x33 51
ss 0x2b 43
ds 0x0 0
es 0x0 0
fs 0x63 99
gs 0x0 0
The disassembly (looks to me like it's faulting at Read+22):
0x00002aaab7b10650 <NET_Read+0>: push %r15
0x00002aaab7b10652 <NET_Read+2>: mov %rdx,%r15
0x00002aaab7b10655 <NET_Read+5>: push %r14
0x00002aaab7b10657 <NET_Read+7>: push %r13
0x00002aaab7b10659 <NET_Read+9>: mov %edi,%r13d
0x00002aaab7b1065c <NET_Read+12>: push %r12
0x00002aaab7b1065e <NET_Read+14>: push %rbp
0x00002aaab7b1065f <NET_Read+15>: push %rbx
0x00002aaab7b10660 <NET_Read+16>: sub $0x38,%rsp
0x00002aaab7b10664 <NET_Read+20>: test %edi,%edi
0x00002aaab7b10666 <NET_Read+22>: mov %rsi,0x8(%rsp)
0x00002aaab7b1066b <NET_Read+27>: js 0x2aaab7b1067c <NET_Read+44>
0x00002aaab7b1066d <NET_Read+29>: lea 1073812(%rip),%rax
# 0x2aaab7c16908 <fdCount>
0x00002aaab7b10674 <NET_Read+36>: cmp (%rax),%edi
0x00002aaab7b10676 <NET_Read+38>: jle 0x2aaab7b1070b <NET_Read+187>
0x00002aaab7b1067c <NET_Read+44>: xor %ebp,%ebp
0x00002aaab7b1067e <NET_Read+46>: test %rbp,%rbp
0x00002aaab7b10681 <NET_Read+49>: je 0x2aaab7b106f9 <NET_Read+169>
0x00002aaab7b10683 <NET_Read+51>: lea 0x10(%rsp),%r14
0x00002aaab7b10688 <NET_Read+56>: callq 0x2aaab7b03dd0 <pthread_self@plt>
0x00002aaab7b1068d <NET_Read+61>: mov %rbp,%rdi
0x00002aaab7b10690 <NET_Read+64>: movl $0x0,0x20(%rsp)
0x00002aaab7b10698 <NET_Read+72>: mov %rax,0x10(%rsp)
0x00002aaab7b1069d <NET_Read+77>: callq 0x2aaab7b03f80
<pthread_mutex_lock@plt>
0x00002aaab7b106a2 <NET_Read+82>: mov %rbp,%rdi
0x00002aaab7b106a5 <NET_Read+85>: mov 0x28(%rbp),%rax
0x00002aaab7b106a9 <NET_Read+89>: mov %rax,0x18(%rsp)
0x00002aaab7b106ae <NET_Read+94>: mov %r14,0x28(%rbp)
0x00002aaab7b106b2 <NET_Read+98>: callq 0x2aaab7b043b0
<pthread_mutex_unlock@plt>
0x00002aaab7b106b7 <NET_Read+103>: mov %r13d,%edi
0x00002aaab7b106ba <NET_Read+106>: mov 0x8(%rsp),%rsi
0x00002aaab7b106bf <NET_Read+111>: xor %ecx,%ecx
0x00002aaab7b106c1 <NET_Read+113>: mov %r15,%rdx
0x00002aaab7b106c4 <NET_Read+116>: callq 0x2aaab7b04160 <recv@plt>
0x00002aaab7b106c9 <NET_Read+121>: mov %rbp,%rdi
0x00002aaab7b106cc <NET_Read+124>: mov %r14,%rsi
0x00002aaab7b106cf <NET_Read+127>: mov %eax,%ebx
0x00002aaab7b106d1 <NET_Read+129>: mov %rax,%r12
0x00002aaab7b106d4 <NET_Read+132>: callq 0x2aaab7b11000 <endOp>
0x00002aaab7b106d9 <NET_Read+137>: inc %ebx
0x00002aaab7b106db <NET_Read+139>: jne 0x2aaab7b106e7 <NET_Read+151>
0x00002aaab7b106dd <NET_Read+141>: callq 0x2aaab7b04380
<__errno_location@plt>
0x00002aaab7b106e2 <NET_Read+146>: cmpl $0x4,(%rax)
0x00002aaab7b106e5 <NET_Read+149>: je 0x2aaab7b10688 <NET_Read+56>
0x00002aaab7b106e7 <NET_Read+151>: mov %r12d,%eax
0x00002aaab7b106ea <NET_Read+154>: add $0x38,%rsp
0x00002aaab7b106ee <NET_Read+158>: pop %rbx
0x00002aaab7b106ef <NET_Read+159>: pop %rbp
0x00002aaab7b106f0 <NET_Read+160>: pop %r12
0x00002aaab7b106f2 <NET_Read+162>: pop %r13
0x00002aaab7b106f4 <NET_Read+164>: pop %r14
0x00002aaab7b106f6 <NET_Read+166>: pop %r15
0x00002aaab7b106f8 <NET_Read+168>: retq
0x00002aaab7b106f9 <NET_Read+169>: callq 0x2aaab7b04380
<__errno_location@plt>
0x00002aaab7b106fe <NET_Read+174>: movl $0x9,(%rax)
0x00002aaab7b10704 <NET_Read+180>: mov $0xffffffff,%eax
0x00002aaab7b10709 <NET_Read+185>: jmp 0x2aaab7b106ea <NET_Read+154>
0x00002aaab7b1070b <NET_Read+187>: movslq %edi,%rax
0x00002aaab7b1070e <NET_Read+190>: lea (%rax,%rax,2),%rbp
0x00002aaab7b10712 <NET_Read+194>: lea 1073639(%rip),%rax
# 0x2aaab7c16900 <fdTable>
0x00002aaab7b10719 <NET_Read+201>: shl $0x4,%rbp
0x00002aaab7b1071d <NET_Read+205>: add (%rax),%rbp
---Type <return> to continue, or q <return> to quit---
0x00002aaab7b10720 <NET_Read+208>: jmpq 0x2aaab7b1067e <NET_Read+46>
0x00002aaab7b10725 <NET_Read+213>: data16
0x00002aaab7b10726 <NET_Read+214>: data16
0x00002aaab7b10727 <NET_Read+215>: data16
0x00002aaab7b10728 <NET_Read+216>: nop
0x00002aaab7b10729 <NET_Read+217>: data16
0x00002aaab7b1072a <NET_Read+218>: data16
0x00002aaab7b1072b <NET_Read+219>: data16
0x00002aaab7b1072c <NET_Read+220>: nop
0x00002aaab7b1072d <NET_Read+221>: data16
0x00002aaab7b1072e <NET_Read+222>: data16
0x00002aaab7b1072f <NET_Read+223>: nop
Looking at the source code for NETRead:
jdk/src/solaris/native/java/net/linux_close.c
*snip*
/*
* Macro to perform a blocking IO operation. Restarts
* automatically if interrupted by signal (other than
* our wakeup signal)
*/
#define BLOCKING_IO_RETURN_INT(FD, FUNC) { \
int ret; \
threadEntry_t self; \
fdEntry_t *fdEntry = getFdEntry(FD); \
if (fdEntry == NULL) { \
errno = EBADF; \
return -1; \
} \
do { \
startOp(fdEntry, &self); \
ret = FUNC; \
endOp(fdEntry, &self); \
} while (ret == -1 && errno == EINTR); \
return ret; \
}
int NET_Read(int s, void* buf, size_t len) {
BLOCKING_IO_RETURN_INT( s, recv(s, buf, len, 0) );
}
Thanks.
492
We had the same issue on our production servers. The problem is basically a stack overflow in the JNI code that is hidden as a segfault: http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7059899
Try jstack to extract a thread dump from the core file produced by the VM and look for infinite loops.
100
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With
