RFR: 8247351: [aarch64] NullPointerException during stack walking (clhsdb "where -a")

[Nick Gasson]

On Fri, 9 Jul 2021 08:37:00 GMT, Nick Gasson <ngasson at openjdk.org> wrote:

> Running the jtreg test serviceability/sa/ClhsdbWhere.java with -Xcomp
> -XX:-TieredCompilation fails with the following exception:
> 
>   Error: java.lang.NullPointerException: Cannot invoke "sun.jvm.hotspot.debugger.Address.getJLongAt(long)" because "valueAddr" is null
>   java.lang.NullPointerException: Cannot invoke "sun.jvm.hotspot.debugger.Address.getJLongAt(long)" because "valueAddr" is null
>           at jdk.hotspot.agent/sun.jvm.hotspot.runtime.CompiledVFrame.createStackValue(CompiledVFrame.java:270)
>           at jdk.hotspot.agent/sun.jvm.hotspot.runtime.CompiledVFrame.getLocals(CompiledVFrame.java:107)
>           at jdk.hotspot.agent/sun.jvm.hotspot.ui.classbrowser.HTMLGenerator.genHTMLForJavaStackTrace(HTMLGenerator.java:1937)
>           at jdk.hotspot.agent/sun.jvm.hotspot.CommandProcessor$43.doit(CommandProcessor.java:1573)
>           at jdk.hotspot.agent/sun.jvm.hotspot.CommandProcessor.executeCommand(CommandProcessor.java:2090)
>           at jdk.hotspot.agent/sun.jvm.hotspot.CommandProcessor.executeCommand(CommandProcessor.java:2060)
> 
> The oop map for the frame being inspected is:
> 
>    ScopeDesc(pc=0x0000ffff8957e000 offset=140):
>    jdk.test.lib.apps.LingeredApp::steadyState at 7 (line 536)
>    Locals
>     - l0: empty
>     - l1: reg rfp [58],oop
>     - l2: empty
>    Monitor stack
>     - @0: monitor{reg rfp [58],oop,stack[16]}
> 
> But RegisterMap::getLocation() returns a null Address for register
> 58 (=RFP).
> 
> This patch fixes two problems: the fp VMReg value used in
> AARCH64Frame::updateMapWithSavedLink should be 29<<1 not 29 because a
> VMReg has two slots in a 64-bit VM.  The other bug is that
> RegisterMap::setLocation and getLocation calculate the long
> locationValid mask using (1 << (i % locationValidTypeSize)) where i and
> locationValidTypeSize are both int.  We need to do this calculation as a
> long if i % locationValidTypeSize can be > 32.  There's no failure on
> x86_64 because that only has 16 integer registers.

OK, understood. Sorry about that.

-------------

PR: https://git.openjdk.java.net/jdk/pull/4737