Using FFI to call JVMTI functions

[David Holmes]

On 26/11/2024 10:48 pm, Jorn Vernee wrote:
> Hello,
> 
> It's correct that FFM upcalls only support callers coming from the 
> 'native' thread state. Looking at the code of 
> IterateOverInstancesOfClass, it holds the heap lock for the duration of 
> the iteration process, so I don't see how we could execute arbitrary 
> Java code during this (through an FFM upcall).
> 
> So, to answer your question: no, you can not use an FFM upcall stub as a 
> callback function for IterateOverInstancesOfClass, or the more modern 
> IterateThroughHeap [1]. In both of these cases, the callback function 
> can not execute Java code.
> 
> Jorn
> 
> [1]: https://docs.oracle.com/en/java/javase/23/docs/specs/ 
> jvmti.html#IterateThroughHeap

We should probably update:

https://docs.oracle.com/en/java/javase/23/docs/specs/jvmti.html#heapCallbacks

to include FFI upcalls as another thing you cannot do during a heap 
callback.

David
-----

> On 26-11-2024 11:59, ght ggg wrote:
>> Hello:
>>
>> I'm trying to use JVMTI's IterateOverInstancesOfClass function to get 
>> all instances of a specified class. The code can be found here: 
>> https://github.com/GGGGGHT/ffitest/blob/main/IterateInstance.java
>>
>> When I call the IterateOverInstancesOfClass method, the JVM crashes. 
>> The crash log is available here: https://github.com/GGGGGHT/ffitest/ 
>> blob/main/hs_err_pid20038.log
>>
>> After examining the code in the UpcallLinker::on_entry method, I found 
>> that this method needs to run on a JavaThread, but the current thread 
>> is a VMThread, which leads to the JVM crash. I'm not entirely sure if 
>> my analysis is correct. If it is correct, does this mean that using 
>> FFI to call functions requiring UPCALL is not feasible? Or are there 
>> alternative approaches to achieve my desired functionality?
>>
>> I look forward to your response.
>> Thanks!
>> GGGGGHT