Double Reporting in GC

[SzpejnaDawid]

Working on port for RISC-V I run some coreFX tests with DOTNET_GCStress=0x4 flag and it results in the explosion of assert in CheckDoubleReporting(...) (path src/coreclr/vm/gcenv.ee.common.cpp). An object from register or stack slot sometimes is doubled. I assume that stackwalker class is correct and REGDISPLAY is updated correctly, so the problem is with GCInfo(?). But could you tell me why an object cannot occur more than once in any slot? Why does this constraint change for pinned objects?

[hez2010]

See #107986, double reporting typical means there're potential GC holes in the implementation.

Double reporting is not allowed unless the reference is pinned, since it would result in incorrect updating of a reference in the slot in case the GC moves the object.

[SzpejnaDawid]

Thx @jkotas, I understand now reasons for it. But what with the situation like that:

void MyMagicalFun(MyClassOfObjects my_obj)
{
    MyClassOfObjects ref1 = my_obj;
    MyClassOfObjects ref2 = my_obj;
    /*
        rest of the code which uses those variables
    */
}

Ryujit can push ref1 and ref2 on the stack and there are at least two stack roots which point to the same object. As I understand even if the above code is correct, they cannot be reported twice when promote = TRUE. So... How is it done that the object is reported only once?

[jkotas]

The double reporting check is concerned about the stack root location, not the object that the stack root points to. The double reporting checks that one stack root location is not reported twice. In your example, there should be 3 stack root locations my_obj, ref1, ref2, each of them should have its own distinct address (assuming no optimizations).

[SzpejnaDawid]

@jkotas, Could tell me whether (in your opinion) my conclusion is correct that double reporting can only be caused by incorrect GCInfo generation if stackwalking and REGDISPLAY are correct? Maybe there is an aspect of GC which i didn't take into consideration.

[jkotas]

Yes, that's correct conclusion. The double reporting issues are typically bugs in GCInfo generation (or in stackwalking - but that's less common).

[SzpejnaDawid]

THX for conversation