<AWT Dev> Review Request For 8146230: Crash in JNI_ArgumentPusherVaArg::JNI_ArgumentPusherVaArg(_jmethodID*, __va_list_tag*)+0xa
ambarish.rapte at oracle.com
Thu Jul 7 13:14:36 UTC 2016
1. The _initIDs() functions are called from static initialization block of particular class, hence these functions would be called when class is getting loaded.
The _initIDs() functions are used to initialize the IDs & get value of Java class members only.
Java_sun_awt_X11_XToolkit_initIDs() initializes IDs and gets value from java side XToolkit class (XToolkit.java)
2. But awtJNI_ThreadYield() is late initialization of IDs. IDs are initialized when first thread has to be yield.
ð Addition of operations to Java_sun_awt_X11_XToolkit_initIDs() is addition to class load time.
ð XToolkit.c::get_xawt_root_shell() is another example of late initialization of jclass, jmethodID.
From: Semyon Sadetsky
Sent: Wednesday, July 06, 2016 11:39 PM
To: Ambarish Rapte; Sergey Bylokhov; Alexander Scherbatiy; Prasanta Sadhukhan; awt-dev at openjdk.java.net
Subject: Re: Review Request For 8146230: Crash in JNI_ArgumentPusherVaArg::JNI_ArgumentPusherVaArg(_jmethodID*, __va_list_tag*)+0xa
Why not simply initialize yieldMethodID separately? For example in Java_sun_awt_X11_XToolkit_initIDs.
On 06.07.2016 14:23, Ambarish Rapte wrote:
Please review the fix for JDK9,
1. Null pointer exception in JNI
The code block was not multi thread safe.
Issue occurs in multi threaded , multi processor environment.
1. Changed the variable used for double checking, to use yieldMethodID .
2. Changed yieldMethodID to volatile.
3. Added AWT_LOCK() over initialize block of code.
4. Removed unrequired err variable.
A drawback of Double Check Locking ( DCL ) is, if the resource assignment is not an atomic operation, The DCL may fail.
But here in the code of concern, is an atomic operation. Hence DCL should work fine.
Please check below reference link for more detailed discussion of DCL with C++.
1. Tested Event tests which pass without any regression of this change.
2. As this change only corrects existing logic, there should be no side effects.
C++ and the Perils of Double-Checked Locking: http://www.aristeia.com/Papers/DDJ_Jul_Aug_2004_revised.pdf
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the awt-dev