RFR: 8392760: Improve liveness/remembered set verification for G1 [v2]

[Kim Barrett]

On Tue, 21 Feb 2023 11:54:15 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:

>> Hi all,
>> 
>>   can I have reviews for this refactoring/cleanup of the liveness/remembered set verification code for G1?
>> 
>> Here is an overview of the changes
>>  - previously liveness and remembered set verification could be done separately. Thismade the code a bit more complicated than necessary; now that remembered set verification always follows liveness verification, the code can assume that and do some minor cleanup. I kept the remembered verification a separate closure (`VerifyRemSetClosure`) because the additional boiler plate code is negligible compared to the other duplication that would have been needed around failure count tracking.
>>  - factored out liveness and remembered set verification from `HeapRegion::verify`, and heavily cleaned up the resulting `HeapRegion::verify_liveness_and_remembered_set`. This includes some removal of code duplication wrt to the failure count tracking.
>> - factored out a special `G1VerificationClosure::is_oop_safe` method that adds more checking of the `Klass`; I hesitated to put this into `Universe::is_oop` as this is much more extensive checking (and I wanted implicit error reporting :))
>> - CollectedHeap::is_oop() should use `klass_raw` because otherwise it will immediately assert on oops with garbage Klass values (the new code will just return `false` as expected). 
>> - some special `oopDesc::print_name_on()` which does not crash on bad oops, similar to existing related debugging helpers.
>> 
>> Thanks,
>>   Thomas
>
> Thomas Schatzl has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains three commits:
> 
>  - Refactoring after 8302975
>  - initial attempts, nothing done
>    
>    further cleanup
>    
>    more cleanup
>    
>    factor out changes that move the verifylivenessclosure
>    
>    fix compilation
>    
>    kbarrett review, indentation
>    
>    Minor removal of unnecessary code
>    
>    initial version
>    
>    more cleanup
>    
>    More improvements
>    
>    More removal of stuff
>    
>    More cleanup
>    
>    More^2 cleanup
>    
>    Remove some changes
>    
>    More cleanups
>    
>    fix compilation
>    
>    Cleanup
>  - removal

src/hotspot/share/gc/g1/heapRegion.cpp line 517:

> 515:       LogStream ls(log.error());
> 516: 
> 517:       MutexLocker x(G1RareEvent_lock, Mutex::_no_safepoint_check_flag);

[pre-existing] It's not clear to me what this lock is protecting.  Similarly for the one in `verify_remset`
(in an unmodified part of it).

src/hotspot/share/gc/g1/heapRegion.cpp line 537:

> 535:       }
> 536:       log.error("----------");
> 537:       return true;

There was a `_num_failures++` here in the old code.  I think it's still needed, even with the bool return.

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

PR: https://git.openjdk.org/jdk/pull/12660