RFR(M): 8030812 : Change the solaris DTrace implementation to use USDT2 instead of USDT1

[Staffan Larsen]

Thanks Coleen!

Yes, removing the !USDT2 macros is the next step. Actually, the first step is to get the Linux system tap implementation to use the USDT2 macros, and then remove the old ones.

/Staffan

On 16 jan 2014, at 19:23, Coleen Phillimore <coleen.phillimore at oracle.com> wrote:

> 
> Staffan,
> This looks good.   The cleanup looks good.  I was hoping that you were going to remove !USDT2 code under those macros.  I assume that's a further change.  I look forward to that.
> Thanks,
> Coleen
> 
> On 01/07/2014 01:24 PM, serguei.spitsyn at oracle.com wrote:
>> On 1/7/14 3:47 AM, Staffan Larsen wrote:
>>> On 20 dec 2013, at 22:07, serguei.spitsyn at oracle.com wrote:
>>> 
>>>> Staffan,
>>>> 
>>>> This is nice change and cleanup.
>>>> I do not see any issues.
>>> Thanks.
>>> 
>>>> The make/bsd/makefiles/dtrace.make shows that the jhelper.d is disabled for bsd
>>>> which means that the jstack action is not supported there.
>>>> I'm not very familiar with the BSD DTrace implementation and wonder if you know any details.
>>>> Does it mean there is no jstack action there, or just more work is needed to make the jhelper.d functional?
>>>> Do we have a bug filed on this?
>>> Short answer: I don’t know anything about jstack or jhelper.d.
>> 
>> The jstack action in a d-script prints mixed java+native stack trace.
>> The jhelper.d is a provider from JVM side (as a part of the DTrace support) that helps
>> the DTrace to translate hexa-decimal pc's to Java symbolic method names.
>> The jhelper.d code is linked into the libjvm.so DOF section where DTrace can find it.
>> 
>> Thanks,
>> Serguei
>> 
>>> 
>>> /Staffan
>>> 
>>>> 
>>>> Thanks,
>>>> Serguei
>>>> 
>>>> 
>>>> On 12/20/13 3:58 AM, Staffan Larsen wrote:
>>>>> The DTrace static probe implementation in Hotspot was written with an earlier version of DTrace. With newer versions, DTrace can create a header file from the contents of the .d file that describes the probes. This newer version (called USDT2) has been used on OS X. Because we have had both versions active, the code has ended up looking ugly because of all the extra macros.
>>>>> 
>>>>> This is a first step in cleaning that up by moving the solaris implementation to use USDT2. The remaining step before USDT1 can be removed is to change the Linux system tap implementation to also use USDT2.
>>>>> 
>>>>> What I have changed is:
>>>>> - Update the solaris dtrace.make to generate the header files. I have used the same code as on bsd.
>>>>> - While I was there, I removed a lot of commented out code from the bsd dtrace.make file.
>>>>> - Updated the hotspot.d files on bsd and solaris so that they have the same contents.
>>>>> - Fixed some compilation errors in compileBroker.cpp with const char*.
>>>>> - May of the USDT2 macro invocations had an extra line break in them. This both looked ugly and confused the solaris compiler, so I removed them. This lead to a _lot_ of changes in jni.cpp - enough changes so that webrev couldn’t handle it, which is why some of the webrev views are broken for this file.
>>>>> -
>>>>> 
>>>>> Testing: I have run the vm.dtrace.testlist on both Solaris and OS X.
>>>>> 
>>>>> webrev: http://cr.openjdk.java.net/~sla/8030812/webrev.00/
>>>>> bug: https://bugs.openjdk.java.net/browse/JDK-8030812
>>>>> 
>>>>> Thanks,
>>>>> /Staffan
>> 
>