RFR: JDK-8308285: Assert on -Xshare:dump when running with -Xlog:cds=trace

[Thomas Stuefe]

On Wed, 17 May 2023 12:17:17 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

> We assert for `java -Xshare:dump -Xlog:cds*=trace`:
> 
> 
> # assert(counter_check_mirror == java_class) failed: must be consistent
> 
> 
> at
> 
> 
>   29 Stack: [0x00007f8821535000,0x00007f8821635000], sp=0x00007f8821632e80, free space=1015k
>   30 Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
>   31 V [libjvm.so+0xda0155] java_lang_Class::primitive_type(oopDesc*)+0x13d (javaClasses.cpp:1367)
>   32 V [libjvm.so+0xd9fb95] java_lang_Class::print_signature(oopDesc*, outputStream*)+0x9f (javaClasses.cpp:1284)
>   33 V [libjvm.so+0xd623f5] InstanceKlass::oop_print_on(oopDesc*, outputStream*)+0x175 (instanceKlass.cpp:3652)
>   34 V [libjvm.so+0x127ec71] oopDesc::print_on(outputStream*) const+0x9f (oop.cpp:47)
>   35 V [libjvm.so+0xd34621] void WalkOopAndArchiveClosure::do_oop_work<narrowOop>(narrowOop*)+0x1e1 (heapShared.cpp:1124)
> ...
> 
> 
> When tracing with at trace level, we print each copied oop in full, including the basic java mirrors. As part of the trace, we print the basic type. The utility function `BasicType java_lang_Class::primitive_type(oop java_class)` does a sanity check: the printed oop is supposed to be a class mirror for a basic type, so it must be the same as the oop resolved via (this oop)->array_klass->element_type->mirror for that type).
> 
> But at this point in the dump process, we have two variants of these basic mirrors: `Universe::_basic_type_mirrors` and `HeapShared::_scratch_basic_type_mirrors`. 
> 
> The patch loosens the assert when dumping such that the assert condition is fulfilled if the mirror is found in either of these two arrays.

@iklam Could you take a look at this please?

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

PR Comment: https://git.openjdk.org/jdk/pull/14031#issuecomment-1551289828