回复: 回复: 回复: [8u] RFR(XS) JFR generate same symbol ID for different symbols (high probabily with CDS)(Internet mail)

[kalinshi(施慧)]

Thanks!

Tests made on linux amd64 with release build
1. jtreg tests, no regression
2.CDS on/off with SPECjbb2015 and SPECjvm with default/profile config. And parse jfr file with Parser to reproduce. (-XX:StartFlightRecording=dumponexit,settings=default/profile,filename=xxx.jfr)

Regards
Hui
-----邮件原件-----
发件人: Andrew Dinn <adinn at redhat.com> 
发送时间: 2020年10月29日 0:37
收件人: kalinshi(施慧) <kalinshi at tencent.com>; jdk8u-dev at openjdk.java.net; Jaroslav Bachorík <jaroslav.bachorik at datadoghq.com>; Langer, Christoph <christoph.langer at sap.com>; Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
主题: Re: 回复: 回复: [8u] RFR(XS) JFR generate same symbol ID for different symbols (high probabily with CDS)(Internet mail)

Hi Hui,

On 19/10/2020 15:07, kalinshi(施慧) wrote:
> Please help review this JFR issue again in 8u. This problem can be easily reproduced with CDS and reproduce steps in JBS.
> 
> Fix includes:
> 1.  Remove map_symbol/map_cstring methods with only hashcode, all lookup must pass symbol or cstring with hashcode.
> 2.  In JfrSymbolId::equals methods, compare symbol reference or cstring contents when hashcodes are same.
> 3.  Rebuild anonymous klass cstring when search its sctring entry id.
> 
> JBS: https://bugs.openjdk.java.net/browse/JDK-8253837
> Webrev: http://cr.openjdk.java.net/~hshi/8253837/webrev_2/
This new patch looks ok to me. However, before accepting it I'd like to hear details of whatever testing you have performed.

regards,


Andrew Dinn
-----------