Hi Steffen,

during startup of the init task, the init thread tries to get an "Env::rm_session" from core. Core responds by giving it a capability with local_name() == 0, which is invalid in my context. The cap is returned from _process.rm_session_cap(), which is implemented in process.h: Rm_session_capability rm_session_cap() const {return _rm.cap(); } _rm is defined as Rm_connection in class Process. I am unsure however, which Constructor belongs to to Rm_connection, and why it sets the capability to zero. Can you help me by pointing me to the right spots?

The 'Env::rm_session' is the RM session managing the address space of the process. It is created by the parent (in your case, core) at construction time of the 'Child' object. The RM session is created via core's 'env()->parent()->session()', which is directed to the core-local 'Core_parent::session()' function (in 'core/main.cc'). For checking if the constuction of init's RM session succeeded, I recommend you to instrument the '_create_session' function in 'core/include/rm_root.h' and the 'Core_parent::session()' function. Check for the capabilities returned by 'ls->root()->session(args)'.

Good luck!

Norman

Hi Steffen,

Genode::Process::Process(Genode::Dataspace_capability, \ Genode::Ram_session_capability, Genode::Cpu_session_capability,\ Genode::Parent_capability, const char*, char* const*):\ name="init" rm_session_cap().cap()=9000\ rm_session_cap().local_name()=16

So after the Process::Process ran, _process->rm_session_cap() should return the correct capabilities... Lets see:

So far, that looks good.

virtual Genode::Session_capability Genode::Child::session(const char*, const char*): _ram_session_cap.cap()=9000 local_name=7

virtual int Genode::Parent_server::dispatch(int, Genode::Ipc_istream&, Genode::Ipc_ostream&): server name="Env::ram_session" args="" cap()=9000 local_name=7

Ok, worked for the ram_session.

virtual Genode::Session_capability Genode::Child::session(const char*, const char*): _cpu_session_cap.cap()=9000 local_name=10

virtual int Genode::Parent_server::dispatch(int, Genode::Ipc_istream&, Genode::Ipc_ostream&): server name="Env::cpu_session" args="" cap()=9000 local_name=10

So the init process is running and uses the RPC interface to the parent to issue session requests to construct its 'env()' object. So init's RM session is definitely working at this point because it is already used to resolve the page faults of init.

Ok, worked for the cpu_session as well.

virtual Genode::Session_capability Genode::Child::session(const char*, const char*): _process.rm_session_cap().cap()=0 local_name=0

virtual int Genode::Parent_server::dispatch(int, Genode::Ipc_istream&, Genode::Ipc_ostream&): server name="Env::rm_session" args="" cap()=0 local_name=0

Did not work for the rm_session. Do you have any ideas?

There should be no difference about how the 'Env::Rm_session' and 'Env::Cpu_session' is handled. I suppose the memory where the process object is stored gets corrupted. Maybe a core-allocator problem? Could you check the page table entry of the memory location of the process object at the creation time of the process object and at the time when the error occurs? It should definitely not change. You could also inspect the memory around the process object to see a possible pattern of memory corruption.

Norman