RFR(S): 8239497: SEGV in EdgeUtils::field_name_symbol(Edge const&)

[Tony Printezis]

Thanks Markus!


—————
Tony Printezis | @TonyPrintezis | tprintezis at twitter.com


On March 17, 2020 at 6:12:46 PM, Markus Gronlund (markus.gronlund at oracle.com)
wrote:

Hi Tony,



That is a good suggestion. I will make the initialization of *modifiers = 0
local to field_name().



Thanks

Markus



*From:* Tony Printezis <tprintezis at twitter.com>
*Sent:* den 17 mars 2020 22:43
*To:* Markus Gronlund <markus.gronlund at oracle.com>;
hotspot-jfr-dev at openjdk.java.net
*Subject:* Re: RFR(S): 8239497: SEGV in EdgeUtils::field_name_symbol(Edge
const&)



Hi Markus,



Thanks for fixing this! I’m not super familiar with the leak profiler code
but the change looks OK to me.



One optional paranoid suggestion: in EdgeUtils::field_name(), *modifiers
will not be set if the method returns NULL. It looks as if it’s set to 0 by
the caller, so it should be OK. But does it make sense to set it to 0 in
field_name() before returning NULL to future-proof this code a bit more?



Tony





—————

Tony Printezis | @TonyPrintezis | tprintezis at twitter.com





On March 17, 2020 at 10:51:11 AM, Markus Gronlund (
markus.gronlund at oracle.com) wrote:

Greetings,

Please help review this change set:

Bug: https://bugs.openjdk.java.net/browse/JDK-8239497
Webrev: http://cr.openjdk.java.net/~mgronlun/8239497/webrev01/
Testing: jdk_jfr
Summary:
To find the proper field offsets for references as part of doing
"path-to-gc-root" traversal, an instance of java/lang/Class must initially
be described using its corresponding metadata, as an Instance(Mirror)Klass.
Only after this can there be an attempt to resolve static fields, which
will then instead use the metadata derived from the InstanceKlass the
java/lang/Class represents. Please see bug for a full description.

Thanks to Tony Printezis for reporting this issue and for analyzing and
testing this change set. Tony will be listed as "Contributed-by".

Thanks
Markus