Hello Bob,
I have an application whose main function creates an object instance based on a Thread. That instance creates another object instance whose main purpose is to read and write to an ffat file system through the libc plug-in for that purpose. As soon as this object attempts to access the file system a error message complaining about no pager object implemented ("/void Genode::Pager_object::unresolved_page_fault_occurred(): not implemented/") occurs.
this message is printed by the base-hw version of core if a program crashes (it corresponds to a segmentation fault on Linux). It would be interesting to see the surrounding log output of the message you quoted. Normally, core prints the instruction pointer and name of the thread that caused the trouble.
On re-reading the Architecture documentation about RM in the Core, it *very* briefly explains that a thread capability for a pager needs to be created for the thread, using an add_client function call. I assumed this would be done in the thread entry() member function and added the following line of code at the the beginning of that function:
Thread_capability ap = Thread::cap(); Pager_capability pc = env()->rm_session()->add_client(ap); env()->cpu_session()->set_pager(ap, pc);
This did not work and I got he same error message about no pager object.
I think you are on the wrong track. For implementing your service, you should not need to deal with the peculiarities of page-fault handling or thread creation at all.
I'm obviously confused about what's going with a process's region manager handling of threads. Could someone please enlighten me?
To get a clearer picture, you may find the documentation useful that I am currently writing. In particular the Sections "Component creation" (in the Architecture chapter) and "Page-fault handling" (in the Under-the-hood chapter) might be of interest:
https://github.com/nfeske/genode-manual
Right now, the document is not finished. So you have to go through the trouble to build it yourself (see the README for the instructions). I hope that it is still useful.
To investigate your problem, may you consider to make your branch along with a run script publicly available (e.g., on GitHub) so that I can reproduce it?
Cheers Norman