Re: Fix corner cases for JIT-enablement

[Daniel H. <dan...@gm...>]

Hi Maynard,

your fixes are looking good for me. Thank you for fixing these problems.

Kind regards,
Daniel Hansel

Maynard Johnson wrote:
> Hi, Daniel,
> A beta user of the JIT enablement for OProfile reported a problem when 
> running SPECjbb2005 with two JVMs on an 8-way POWER6 server. 
> Occasionally, the summary report would show some number of samples 
> attributed to two *.jo files (corresponding to the two JVMs); but when 
> examining the detailed report (with '--symbols'), only one of the *.jo 
> files showed up with samples attributed to it.
>
> I was able to reproduce the symptom by stopping oprofile shortly after 
> starting the java app.  Actually, depending on the timing of my 
> starting and stopping of the java app and oprofile, I was able to 
> reproduce a couple different symptoms:
>  o Sometimes, a jit dumpfile would be produced, but incomplete
>    information would cause an abort of opjitconv before even
>    creating a .jo file.  The opcontrol script that calls opjitconv
>    would then blindly try to mv/chmod the .jo file that was not
>    created, causing some error messages.
>  o Other times, opjitconv wouldn't find enough useful information the
>    jit dumpfile.  A .jo file would be produced, but no symbol
>    information would be found. (This caused the symptom the beta user
>    reported to me above.)
>
> The attached patch addresses both of these problems.
>
> Unless anyone sees something wrong with the patch, I'll commit it to 
> the JIT_SUPPORT branch tomorrow.
>
> Regards,
> -Maynard