On 4/7/15 2:01 PM, [email protected] wrote:
On 4/7/15 11:20 AM, Coleen Phillimore wrote:
шI see. The PreviousVersionWalker is gone in 9 and you're using it
to find the method.
But this is only going to find the method if it happens to be in the
cpCache somewhere. I think
pv_node->prev_constant_pool()->pool_holder() is scratch_class (not
100% sure). In which case, you can look at that class to find the
method with idnum.
I'm not 100% sure either.
Ok, I will check if using the prev_constant_pool()->pool_holder() is
going to work.
It'd make it more consistent with the jdk 9 fix.
This is the updated webrev.
It is more consistent with the jdk 9 fix.
Thank you for the suggestion!
Thanks,
Serguei
Important part is that at least the saved cpref can be used to find
method name in the
the_class cpool if the attempt to find in the previous list cpCaches
was unsuccessful.
Thanks,
Serguei
Coleen
On 4/7/15, 1:09 AM, [email protected] wrote:
Coleen,
Thank you for looking at this!
The main difference is in instanceKlass.cpp and instanceKlass.hpp.
8u60:
+Method* InstanceKlass::method_with_orig_idnum(int idnum) {
+ if (idnum >= methods()->length()) {
+ return NULL;
+ }
+ Method* m = methods()->at(idnum);
+ if (m != NULL && m->orig_method_idnum() == idnum) {
+ return m;
+ }
+ // Obsolete method idnum does not match the original idnum
+ for (int index = 0; index < methods()->length(); ++index) {
+ m = methods()->at(index);
+ if (m->orig_method_idnum() == idnum) {
+ return m;
+ }
+ }
+ // None found, return null for the caller to handle.
+ return NULL;
+}
+
+
+Method* InstanceKlass::method_with_orig_idnum(int idnum, int version) {
+ if (constants()->version() == version) {
+ return method_with_orig_idnum(idnum);
+ }
+ ConstantPoolCache* cp_cache = NULL;
+ PreviousVersionWalker pvw(Thread::current(), (InstanceKlass*)this);
+ for (PreviousVersionNode * pv_node = pvw.next_previous_version();
+ pv_node != NULL; pv_node = pvw.next_previous_version()) {
+ ConstantPool* prev_cp = pv_node->prev_constant_pool();
+ if (prev_cp->version() == version) {
+ cp_cache = prev_cp->cache();
+ break;
+ }
+ }
+ if (cp_cache == NULL) {
+ return NULL; // The pv_node is gone, no method is found
+ }
+ Method* method = NULL;
+ for (int i = 0; i < cp_cache->length(); i++) {
+ ConstantPoolCacheEntry* entry = cp_cache->entry_at(i);
+ method = entry->get_interesting_method_entry(this);
+ if (method == NULL) {
+ continue; // skip uninteresting entries
+ }
+ if (method->orig_method_idnum() == idnum) {
+ return method;
+ }
+ }
+ // None found, return null for the caller to handle.
+ return NULL;
+}
jdk 9:
+Method* InstanceKlass::method_with_orig_idnum(int idnum) {
+ if (idnum >= methods()->length()) {
+ return NULL;
+ }
+ Method* m = methods()->at(idnum);
+ if (m != NULL && m->orig_method_idnum() == idnum) {
+ return m;
+ }
+ // Obsolete method idnum does not match the original idnum
+ for (int index = 0; index < methods()->length(); ++index) {
+ m = methods()->at(index);
+ if (m->orig_method_idnum() == idnum) {
+ return m;
+ }
+ }
+ // None found, return null for the caller to handle.
+ return NULL;
+}
+
+
+Method* InstanceKlass::method_with_orig_idnum(int idnum, int version) {
+ InstanceKlass* holder = get_klass_version(version);
+ if (holder == NULL) {
+ return NULL; // The version of klass is gone, no method is found
+ }
+ Method* method = holder->method_with_orig_idnum(idnum);
+ return method;
+}
Thanks,
Serguei
On 4/6/15 6:49 PM, Coleen Phillimore wrote:
This backport looks good. Why didn't the patch apply cleanly? It
seems mostly the same to me.
Thanks,
Coleen
On 4/6/15, 7:50 PM, [email protected] wrote:
Coleen and Dan,
I'm asking you as the jdk 9 reviewers...
Please, let me know if you have any chance and plans to review this.
Would it be enough to have just one (R) reviewer for a non-clean
backport?
Thanks,
Serguei
On 4/2/15 5:03 PM, [email protected] wrote:
Please, review the jdk 8u60 backport for:
https://bugs.openjdk.java.net/browse/JDK-8067662
8u60 hotspot webrev:
http://cr.openjdk.java.net/~sspitsyn/webrevs/2015/hotspot/8067662-JVMTI-trace.8u60
8u60 webrev for unit test update (the fix applies cleanly):
http://cr.openjdk.java.net/~sspitsyn/webrevs/2015/jdk/8067662-JVMTI-trace-test.8u60
9 hotspot webrev:
http://cr.openjdk.java.net/~sspitsyn/webrevs/2015/hotspot/8067662-JVMTI-trace.2
9 webrev for unit test update:
http://cr.openjdk.java.net/~sspitsyn/webrevs/2015/jdk/8067662-JVMTI-trace-test.2
Summary:
The 9 fix can not be applied cleanly so that I have to rework
the fix a little bit.
An NPE is trown in a thread dumping via JMX when doing CPU
profiling in NetBeans Profiler and VisualVM.
The issue is that the methods and related klass versions are
not kept alive if a class redefinition
takes place between catching a Throwable and taking a thread dump.
It can result in loosing an obsolete method, and so, the
reconstruction of method name becomes impossible.
In such a case, the null reference is returned instead of a
valid method name.
The solution is to use current klass version and method
orig_idnum instead of ordinary method idnum
to find required method pointer. In the worst case when the
related klass version is gone
(was not included to or was removed from the previous_versions
linked list), a saved method name
CP index of the latest klass version can be used to restore the
method name.
The footprint extra overhead for this approach is u2 per stack
frame.
Testing in progress:
In progress: nsk redefine classes tests, JTREG
java/lang/instrument
Thanks,
Serguei
