[dtrace-discuss] plockstat/dtrace core dump S10U3

[Robert Harris]

Hi Stefan,

Stefan Parvu wrote:

> Im able to reproduce a crash from plockstat everytime Im tracing a JVM pid. I do recall a problem related with this one, but not clear if this has been fixed in U3 or is it planned for U4 ? Any BugID opened for this stack trace:
> 
>> ::stack
> libc.so.1`strlen+0x50(100003faa, ffffffff7ffff5c8, ffffffff7eca1114, ffffffff7fffec79, 0, 100003fa9)
> libc.so.1`snprintf+0x88(ffffffff7ffff8f0, 0, 100003fa8, 0, ffffffff7ffff620, 100003000)
> 0x100001cd8(10068c060, fecb8d70, ffffffff7ffff8f0, 28, 1, 1d)
> 0x100001ff4(100805b08, 100805b28, 5, 100805b20, 100805af8, 100106990)
> libdtrace.so.1`dt_aggregate_walk_sorted+0xa4(10010b500, 100001de8, 0, ffffffff7f124f08, 1, 100806e70)
> main+0xad0(100001, 100106000, 100000, 100001000, 100107228, 100106)
> _start+0x17c(0, 0, 0, 0, 0, 0)
> 
> the command was: plockstat -a -p <jvm_pid>

This is bug 6296612, which IIRC is fixed in the latest Solaris Developer 
Express. I expect to have it released as a patch in due course. In the 
mean time, you should be able to prevent the bug from occuring by 
disabling large pages with

	echo segvn_lpg_disable/W 1 | mdb -kw

This will possibly have a performance impact; large pages can be 
reenabled with

	echo segvn_lpg_disable/W 0 | mdb -kw

If you have an urgent need for a better temporary fix on Solaris 10 and 
you have a service contract then please log a case and ask for an IDR.

Robert