Hi Collen,
Thank you for making this update!
It looks good to me.
One nit:
http://cr.openjdk.java.net/~coleenp/2019/8212160.03/webrev/test/hotspot/jtreg/serviceability/jvmti/CompiledMethodLoad/libCompiledZombie.cpp.html
46 // Continuously generate CompiledMethodLoad events
for all currently compiled methods
47 void JNICALL GenerateEventsThread(jvmtiEnv* jvmti,
JNIEnv* jni, void* arg) {
48 jvmti->SetEventNotificationMode(JVMTI_ENABLE,
JVMTI_EVENT_COMPILED_METHOD_LOAD, NULL);
49 int count = 0;
50
51 while (true) {
52 events = 0;
53
jvmti->GenerateEvents(JVMTI_EVENT_COMPILED_METHOD_LOAD);
54 if (events != 0 && ++count == 200) {
55 printf("Generated %d events\n", events);
56 count = 0;
57 }
58 }
59 }
The above can be simplified a little bit:
if (events % 200 == 199) {
printf("Generated %d events\n", events);
}
Then this line is not needed too:
49 int count = 0;
I answered this too fast. There are two conditions where I
want this to not print. First is where events == 0 and the
other for every 200 events that are non-zero.
I could use if (events != 0 && count++ % 200), but I
thought what I had makes more sense and I don't have to worry
about when ++ happens.
Then you could replace it with:
if (events % 200 == 0) {
But that would still print when events == 0, which I don't want.
If I print them all for the little test case, it's ok, but when I
run this with Swingset2, it's too much output. I only want to see
a few lines for this:
----------System.out:(3/113)----------
Test passes if it doesn't crash while posting compiled method
events.
Generated 285 events
Generated 1002 events
----------System.err:(1/15)----------
The count is the number of times through the GenerateEvents loop,
which resets events to zero each time, then prints the number of
events for every 200 times through the GenerateEvents loop. So I
need both count and events.
Coleen
But it is up to you. :)
Thanks,
Serguei
Thanks,
Coleen
Thanks Dan. I moved the field. For some reason I thought
that class did more/different things than hold per-thread
information.
I've retested this version with tiers 2-6.
incr webrev at http://cr.openjdk.java.net/~coleenp/2019/8212160.03.incr/webrev
full webrev at http://cr.openjdk.java.net/~coleenp/2019/8212160.03/webrev
Thanks to Serguei for offline discussion.
Coleen
On 12/4/19 7:40 PM, Daniel D.
Daugherty wrote:
Generally speaking, JVM/TI related things should be
in JvmtiThreadState
instead of directly in the Thread class. That way the
extra space is only
consumed when JVM/TI is in use and only when a Thread
does something that
requires a JvmtiThreadState to be created.
Please reconsider moving _jvmti_event_queue.
Dan
Hi Serguei,
Hi Collen, (no problem)
It looks good in general.
Thank you a lot for sorting this out!
Just a couple of comments.
http://cr.openjdk.java.net/~coleenp/2019/8212160.01/webrev/src/hotspot/share/runtime/thread.hpp.frames.html
1993 protected:
1994 // Jvmti Events that cannot be posted in their current context.
1995 // ServiceThread uses this to collect deferred events from NonJava threads
1996 // that cannot post events.
1997 JvmtiDeferredEventQueue* _jvmti_event_queue;
As David I also have a concern about footprint of
having the _jvmti_event_queue
field in the Thread class.
I'm thinking if it'd be better to move this field
into the JvmtiThreadState class.
Please, see jvmti_thread_state() and
JvmtiThreadState::state_for(JavaThread *thread).
The reason I have it directly in JavaThread is so that
the GC oops_do and nmethods_do code can find it
easily. I like your idea of hiding it in jvmti but
this doesn't seem good to have this code know about
jvmtiThreadState, which seems to be a queue of Jvmti
states. I also don't want to have jvmtiThreadState to
have to add an oops_do() or nmethods_do() either.
I don't envision an extension for this function but I
do for JvmtiDeferredEventQueue::post(). I have a
small enhancement that would handoff the entire queue
to the ServiceThread and have it call post() to post
all the events rather than one at a time.
So I'll rename this one
post_compiled_method_load_event() and leave the other
post() as is for now.
open webrev at http://cr.openjdk.java.net/~coleenp/2019/8212160.02.incr/webrev
Thanks,
Coleen
Summary:
Add local deferred event list to thread to post
events outside CodeCache_lock.
This patch builds on the patch for JDK-8173361.
With this patch, I made the
JvmtiDeferredEventQueue an instance class (not
AllStatic) and have one per thread. The CodeBlob
event that used to drop the CodeCache_lock and
raced with the sweeper thread, adds the events it
wants to post to its thread local list, and
processes it outside the lock. The list is walked
in GC and by the sweeper to keep the nmethods from
being unloaded and zombied, respectively.
Also, the jmethod_id field in nmethod was only
used as a boolean so don't create a jmethod_id
until needed for post_compiled_method_unload.
Ran hs tier1-8 on linux-x64-debug and the stress
test that crashed in the original bug report.
open webrev at http://cr.openjdk.java.net/~coleenp/2019/8212160.01/webrev
bug link https://bugs.openjdk.java.net/browse/JDK-8212160
Thanks,
Coleen
|
