RFR(S): 8241638: launcher time metrics alway report 1 on Linux when _JAVA_LAUNCHER_DEBUG set(Internet mail)

Mon Apr 6 02:37:41 UTC 2020

On 6/04/2020 12:20 pm, Henry Jen wrote:
> On Apr 5, 2020, at 7:15 PM, David Holmes <david.holmes at oracle.com> wrote:
>>
>> On 6/04/2020 11:50 am, David Holmes wrote:
>>> There is something not right here ...
>>> On 4/04/2020 3:13 pm, Henry Jen wrote:
>>>> Internal test shows that inline implementation is not working for some Solaris artifacts, because the -DHAVE_GETHRTIME is not consistently defined, so it is actually broken. :)
>>>>
>>>>> [2020-04-03T15:59:26,981Z] Creating support/test/hotspot/jtreg/native/bin/jvm-test-launcher from 1 file(s)
>>> Are you sure that line actually pertains to any error? The test defines a custom launcher which doesn't use libjli so should never be including the header file we are discussing.
>>>>> [2020-04-03T16:02:10,984Z]
>>>>> [2020-04-03T16:02:10,984Z] ERROR: Build failed for target 'default (product-bundles test-bundles static-libs-bundles)' in configuration 'solaris-sparcv9-open' (exit code 2)
>>>>> [2020-04-03T16:02:11,051Z]
>>>>> [2020-04-03T16:02:11,051Z] === Output from failing command(s) repeated here ===
>>>>> [2020-04-03T16:02:11,055Z] * For target support_native_java.base_libjli_BUILD_LIBJLI_link:
>>>>> [2020-04-03T16:02:11,061Z] Undefined            first referenced
>>>>> [2020-04-03T16:02:11,061Z]  symbol                  in file
>>>>> [2020-04-03T16:02:11,061Z] getTimeMicros                       /export/home/opt/mach5/mesos/work_dir/e101deec-9613-4d46-a64d-559e689496aa/workspace/build/solaris-sparcv9-open/support/native/java.base/libjli/java.o
>>>>> [2020-04-03T16:02:11,061Z] ld: fatal: symbol referencing errors
>>> This looks like a direct linkage error. AFAICS ./launcher/LauncherCommon.gmk only defines -DHAVE_GETHRTIME for launcher executables. But if we are building libjli it is not an executable. I'm suspecting there is actually a long standing build bug here from when libjli was introduced. Possibly only evident on an incremental build.
>>
>> I can confirm that the flags set in LauncherCommon.gmk are not passed to the compilation of java_md_solinux.c (I added a custom -DDAVIDH to the linux flags and checked the build). So I have no idea how this has been working, if indeed it actually has.
>>
> 
> I should say it’s the inconsistency for building java.c for launcher executable and libjli.so, thus cause libjli.so failed to build. It wasn’t detected before because CounterGet is defined as no-op, so nothing to link with.

My point is that this seems completely broken. HAVE_GETHRTIME is never 
defined when building libjli, only when building the launcher 
executables, and the launcher executable code never calls CounterGet(). 
This suggests that the launcher debug code on Solaris has been using the 
same code as linux and thus should be seen to be reporting the same 
thing ... which it is ...

_JAVA_LAUNCHER_DEBUG=true 
/java/re/jdk/9/promoted/latest/binaries/solaris-x64/bin/java -version
----_JAVA_LAUNCHER_DEBUG----
Launcher state:
         First application arg index: -1
         debug:on
         javargs:off
         program name:java
         launcher name:java
         javaw:off
         fullversion:9+181
Command line args:
argv[0] = /java/re/jdk/9/promoted/latest/binaries/solaris-x64/bin/java
argv[1] = -version
JRE path is .../java_re/jdk/9/fcs/181/binaries/solaris-x64
jvm.cfg[0] = ->-server<-
jvm.cfg[1] = ->-client<-
1 micro seconds to parse jvm.cfg
Default VM: server
Does 
`.../export/java_re/jdk/9/fcs/181/binaries/solaris-x64/lib/server/libjvm.so' 
exist ... yes.
mustsetenv: FALSE
JVM path is 
.../export/java_re/jdk/9/fcs/181/binaries/solaris-x64/lib/server/libjvm.so
1 micro seconds to LoadJavaVM

Always reports 1 microsecond just like Linux.

This whole setup is completely broken at the build level.

Cheers,
David

> Cheers,
> Henry
> 
> 
>> David
>>
>>> David
>>> -----
>>>>> [2020-04-03T16:02:11,082Z]
>>>>> [2020-04-03T16:02:11,082Z] * All command lines available in /export/home/opt/mach5/mesos/work_dir/e101deec-9613-4d46-a64d-559e689496aa/workspace/build/solaris-sparcv9-open/make-support/failure-logs.
>>>>> [2020-04-03T16:02:11,086Z] === End of repeated output ===
>>>>> [2020-04-03T16:02:11,094Z]
>>>>> [2020-04-03T16:02:11,094Z] === Make failed targets repeated here ===
>>>>> [2020-04-03T16:02:11,736Z] CoreLibraries.gmk:206: recipe for target '/export/home/opt/mach5/mesos/work_dir/e101deec-9613-4d46-a64d-559e689496aa/workspace/build/solaris-sparcv9-open/support/modules_libs/java.base/libjli.so' failed
>>>>> [2020-04-03T16:02:11,737Z] make/Main.gmk:195: recipe for target 'java.base-libs' failed
>>>>> [2020-04-03T16:02:11,739Z] === End of repeated output ===
>>>>> [2020-04-03T16:02:11,741Z]
>>>>
>>>>
>>>> I verified that either move implementation into .c as a function body[1] or change to #ifdef __solaris__[2] will fix that. So I think we will change to detect __solaris__ as webrev[2] rather than have an extra #define. If some other build want to have that, they can be modify that #ifdef easily.
>>>>
>>>> As I look into it, I found Mac have similar implementation with minor mistake, so I fixed that as well. Please review following based on zhanglin’s patch.
>>>>
>>>> I’ll push [2] once I got a +1.
>>>>
>>>> [1] http://cr.openjdk.java.net/~henryjen/jdk/8241638.0/webrev/
>>>> [2] http://cr.openjdk.java.net/~henryjen/jdk/8241638.1/webrev/
>>>>
>>>> Cheers,
>>>> Henry
>>>>
>>>>> On Apr 2, 2020, at 6:17 AM, Alan Bateman <Alan.Bateman at oracle.com> wrote:
>>>>>
>>>>> On 02/04/2020 11:26, linzang(臧琳) wrote:
>>>>>> :
>>>>>>              Here is the updated webrev : http://cr.openjdk.java.net/~lzang/8241638/webrev04/
>>>>>>              Thanks for your help!
>>>>>>
>>>>> webrev04 looks good. My preference was to was to replace #ifdef HAVE_GETHRTIME with #ifdef __solaris__ but there doesn't seem to be appetite to do this now. I think Henry has offered to help sponsor.
>>>>>
>>>>> -Alan.
> 