RFR(S): 8241638: launcher time metrics alway report 1 on Linux when _JAVA_LAUNCHER_DEBUG set(Internet mail)
linzang(臧琳)
linzang at tencent.com
Mon Apr 6 00:53:04 UTC 2020
> Certainly we can change to not inline by combining both fixes. I don’t think inline is necessary causing problem but that means header and the binary does need to match.
How about use macro instead of function?
+#define getTimeMicros() ({ \
+ uint64_t result = 0; \
+ struct timeval tv; \
+ if (gettimeofday(&tv, NULL) != -1) { \
+ result = 1000000LL * (uint64_t)tv.tv_sec; \
+ result += (uint64_t)tv.tv_usec; \
+ } \
+ result; \
+})
I have tested it works on linux, not sure whether it is ok for solaris.
BRs,
Lin
On 2020/4/6, 3:37 AM, "Henry Jen" <henry.jen at oracle.com> wrote:
> On Apr 5, 2020, at 6:52 AM, Alan Bateman <Alan.Bateman at oracle.com> wrote:
>
>
> On 05/04/2020 14:17, David Holmes wrote:
>> 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. :)
>>
>> The problem is in defining that function as inline rather than the -DHAVE_GETHRTIME.
>>
>>>> [2020-04-03T15:59:26,981Z] Creating support/test/hotspot/jtreg/native/bin/jvm-test-launcher from 1 file(s)
>>
>> The build rules for this special test launcher need to be examined as something seems wrong to me.
> I assume it's just that it's just compiled differently to the regular launchers and just not noticed that it was missing -DHAVE_GETHRTIME (unless for anyone to use it with _JAVA_LAUNCHER_DEBUG set).
This is my understanding as well, we built something without define -DHAVA_GETHRTIME but linked with the library that was built with that defined and use inline function. We won’t notice this before because CounterGet is no-op before.
> If we go with Henry's second webrev then I assume -DHAVE_GETHRTIME can be dropped from LauncherCommon.gmk to avoid confusing any further maintainers in this area.
Right, I can drop that as well.
> Also is there a strong need for the non-Solaris getTimeMicros to have be inline?
>
Certainly we can change to not inline by combining both fixes. I don’t think inline is necessary causing problem but that means header and the binary does need to match.
Cheers,
Henry
More information about the core-libs-dev
mailing list