Genode 16.05 Fault handler

Sebastian Sumpf Sebastian.Sumpf at ...1...
Mon Sep 5 19:04:34 CEST 2016


He Denis,

On 09/03/2016 12:03 PM, Denis Huber wrote:
> Hello Sebastian,
> 
> I think, I have expressed myself incorrectly. I don't want to handle the 
> page faults with the same thread which also causes them. I want to find 
> out the usage of the new Signal_handler API: I want to create a simple 
> component which causes a page fault and which also registers a page 
> fault handler for its address space, which wil resolve the page faults 
> in a separate Entrypoint (= separate Thread).
> 
> Practically, I have created a test scenario [1] with the old 
> Signal_receiver and Signal_context syntax, where a component registers a 
> page fault handler for its address space and then causes a page fault.
> 
> [1] 
> https://github.com/702nADOS/genode-CheckpointRestore-SharedMemory/blob/7c1c6183a5132aa67d2604221d4eb6ee39106c64/src/test/fault_handler_old/main.cc
> 
> Now, I want to change this scenario to use the new Signal_handler 
> syntax, where a component registers a Signal_handler and then causes a 
> page fault, which is handled by the Signal_handler.
> 
> My understanding of a Signal_handler is, that the Entrypoint, which is 
> passed at the creation of a Signal_handler, catches a page fault signal 
> from core and calls the function passed to the Signal_handler. Because 
> an Entrypoint is basically a thread, it will do the same as 
> Local_fault_handler class in [1] (Is it correct?).
> 
> Based on that thoughts I created this component [2].
> 
> [2] 
> https://github.com/702nADOS/genode-CheckpointRestore-SharedMemory/blob/81f6a705b5b57bc992c47566747a8273fb056c33/src/test/fault_handler/main.cc
> 
> I handle the "hen and egg" problem with a separate Entrypoint which is 
> actually another thread (other then the thread which causes the page 
> fault). But my solution does not work. Where do I have a 
> misunderstanding of the Signal_handler/Entrypoint concept?
> 
> Basically, I want to know how to handle page faults using a 
> Signal_handler object. In my project I will use a parent component, 
> which will create several Region_maps and provide one Signal_handler for 
> all those Region_maps. One Region_map is given to the child as a 
> dataspace, when it allocates ram from its ram_session. The Region_map is 
> initially empty. When the child accesses an address in that Region_map, 
> the caused page fault will be caught by the Signal_handler and a 
> dataspace, backed with physical memory, attached.

You seem to be doing it all right. I have updated the recent fault
handler example to use a second entrypoint
(https://github.com/ssumpf/genode/commit/ff23c7227944324b55fc281f26b8e5bfc12e14fa).
Doing this, I experienced the same problems as yourself. Namely, that
the page-fault signal is not delivered. This is caused by the fact of
the signal proxy thread of the entrypoint (that delivers the signal) not
being started. I fixed this and it works for now. It might just be a
bug, but I will discuss this issue tomorrow.

Regards,

Sebastian

> Thank you for your help in advance ^^
> 
> 
> Kind regards,
> Denis
> 
> On 31.08.2016 18:23, Sebastian Sumpf wrote:
>> Hi Denis,
>>
>> On 08/31/2016 02:24 PM, Denis Huber wrote:
>>> Hello Sebastian,
>>>
>>> thank you for the working solution :)
>>>
>>> Now the page faults of threads other than the main-thread can be
>>> handled. What about the main-thread? How can I change the example to be
>>> able to handle page faults from the main-thread?
>>>
>>> Do I have to create an extra thread which is dedicated to handle page
>>> faults of the other threads like the old Signal_receiver/Signal_context
>>> approach [1]? I thought the new design of the Entrypoint provides this
>>> feature (Entrypoint handles page faults in its thread). Or is my
>>> understanding wrong?
>>
>> as I said yesterday, a thread (what an EP essentially is) cannot handle
>> its own page fault. It is a hen and egg problem. Microkernels just save
>> the thread state and call a pager thread in order to resolve the page
>> fault. Note, that this thread does not have to be within the same
>> program (or protection domain). What I don't understand, why would you
>> want a self paging program? The memory has to be provided and paged from
>> the outside anyway (in Genode that would be the parent or core)?
>>
>> Cheers,
>>
>> Sebastian
>>
>>>
>>> [1]
>>> https://github.com/702nADOS/genode-CheckpointRestore-SharedMemory/blob/7c1c6183a5132aa67d2604221d4eb6ee39106c64/src/test/fault_handler_old/main.cc
>>>
>>>
>>> Kind regards,
>>> Denis
>>>
>>> On 30.08.2016 18:36, Sebastian Sumpf wrote:
>>>> Hi Denis,
>>>>
>>>> On 08/30/2016 04:21 PM, Denis Huber wrote:
>>>>> Hello Sebastian,
>>>>>
>>>>> I created a new thread which causes the page fault, and also assigned an
>>>>> extra Entrypoint for the Signal_handler [1], but still no success.
>>>>> Perhaps I am using the Signal_handler in a wrong way.
>>>>>
>>>>> [1]
>>>>> https://github.com/702nADOS/genode-CheckpointRestore-SharedMemory/blob/c58b4144bff720f837112c590ebd2148e9c3f199/src/test/fault_handler/main.cc
>>>>>
>>>>
>>>> You are mixing some things up here, I pushed a working version of your
>>>> code (tested on Fisco.OC in Qemu) to GitHub:
>>>>
>>>> https://github.com/ssumpf/genode/commit/0075a0026faa8079f03dfb4d43ec281ef51f1083
>>>>
>>>> I hope this clarifies things a little. The relevant output should look
>>>> like this:
>>>>
>>>>> [init -> pf-signal_handler] --- pf-signal_handler started ---
>>>>> no RM attachment (WRITE pf_addr=1000 pf_ip=10001d9 from 2d2000)
>>>>> Warning: could not resolve pf=0x1000 ip=0x10001d9
>>>>> [init -> pf-signal_handler]   Region map state is WRITE_FAULT pf_addr=0x1000
>>>>> [init -> pf-signal_handler]   Allocating dataspace and attaching it to the region map
>>>>> [init -> pf-signal_handler] --- pf-signal_handler ended ---
>>>>
>>>> Sebastian
>>>>
>>>> P.S. As I see it, you needed the 'join' because the thread was created
>>>> on the stack again ;) One has to use the heap or bss/data.
>>>>
>>>>
>>>>> Kinds regards,
>>>>> Denis
>>>>>
>>>>> On 30.08.2016 15:45, Sebastian Sumpf wrote:
>>>>>> Hello again,
>>>>>>
>>>>>> I missed that part
>>>>>>
>>>>>>> *((unsigned int*)0x1000) = 1;
>>>>>>
>>>>>> in my last reply. You cannot cause a page fault in your entry point and
>>>>>> also handle it in the entry point thread, because the entry point is not
>>>>>> able to receive signals when it is in page fault. It is a hen and egg
>>>>>> problem.
>>>>>>
>>>>>> If you cause the fault from another Genode thread, it should work as
>>>>>> expected.
>>>>>>
>>>>>> Sebastian
>>>>>>
>>>>>>
>>>>>> On 08/30/2016 03:07 PM, Denis Huber wrote:
>>>>>>> Hello Sebastian,
>>>>>>>
>>>>>>> I moved sigh to the member attributes [1], but without success. I also
>>>>>>> tried to use a heap to store the Signal_handler, also without success.
>>>>>>>
>>>>>>> [1]
>>>>>>> https://github.com/702nADOS/genode-CheckpointRestore-SharedMemory/blob/93eaec436ce9b6d3b3a6663558bff9af925ae70f/src/test/fault_handler/main.cc#L19
>>>>>>>
>>>>>>> I am using run script [2], if it helps.
>>>>>>>
>>>>>>> [2]
>>>>>>> https://github.com/702nADOS/genode-CheckpointRestore-SharedMemory/blob/93eaec436ce9b6d3b3a6663558bff9af925ae70f/run/fault_handler.run
>>>>>>>
>>>>>>>
>>>>>>> Kind regards,
>>>>>>> Denis
>>>>>>>
>>>>>>> On 30.08.2016 14:36, Sebastian Sumpf wrote:
>>>>>>>> Hi Denis,
>>>>>>>>
>>>>>>>>> Signal_handler<Main> sigh = {env.ep(), *this, &Main::_handle_fault};
>>>>>>>>
>>>>>>>> The line above, creates the Signal_handler on the stack and destructs
>>>>>>>> this handler as soon as the 'Main' constructor is finished.
>>>>>>>>
>>>>>>>> Therefore, 'sigh' should be a member of the 'Main' class.
>>>>>>>>
>>>>>>>> Regards,
>>>>>>>>
>>>>>>>> Sebastian
>>>>>>>>
>>>>>>>> On 08/30/2016 02:10 PM, Denis Huber wrote:
>>>>>>>>> Dear Genode community,
>>>>>>>>>
>>>>>>>>> in Genode 16.05 the old usage of Signal_receivers and Signal_contexts is
>>>>>>>>> considered deprecated. Instead Signal_handler and Signal_transmitter
>>>>>>>>> shall be used.
>>>>>>>>>
>>>>>>>>> I am trying to use a Signal_handler as a fault handler for env's address
>>>>>>>>> space. I wrote a short test to illustrate the problem [1].
>>>>>>>>>
>>>>>>>>> [1]
>>>>>>>>> https://github.com/702nADOS/genode-CheckpointRestore-SharedMemory/blob/c9a5a1f999a2ce1c6ffd5c316eee127c93d87308/src/test/fault_handler/main.cc
>>>>>>>>>
>>>>>>>>> The output of [1] is as follows:
>>>>>>>>> "
>>>>>>>>> [init -> pf-signal_handler] --- pf-signal_handler started ---
>>>>>>>>> no RM attachment (WRITE pf_addr=1000 pf_ip=100061c from 2ae000)
>>>>>>>>> Genode::Pager_entrypoint::entry()::<lambda(Genode::Pager_object*)>:
>>>>>>>>> Could not resolve pf=1000 ip=100061c
>>>>>>>>> "
>>>>>>>>>
>>>>>>>>> The signal handler was registered correctly, because the line
>>>>>>>>> "
>>>>>>>>> Genode::Core_pd_session_component::submit(Genode::Capability<Genode::Signal_context>,
>>>>>>>>> unsigned int)::<lambda(Genode::Signal_context_component*)>: invalid
>>>>>>>>> signal-context capability
>>>>>>>>> "
>>>>>>>>> is missing. However, the method _handle_fault() is never entered. What
>>>>>>>>> am I doing wrong?
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Kind regards,
>>>>>>>>> Denis
>>>>>>>>>
>>>>>>>>> ------------------------------------------------------------------------------
>>>>>>>>> _______________________________________________
>>>>>>>>> genode-main mailing list
>>>>>>>>> genode-main at lists.sourceforge.net
>>>>>>>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ------------------------------------------------------------------------------
>>>>>>>> _______________________________________________
>>>>>>>> genode-main mailing list
>>>>>>>> genode-main at lists.sourceforge.net
>>>>>>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>>>>>>
>>>>>>>
>>>>>>> ------------------------------------------------------------------------------
>>>>>>> _______________________________________________
>>>>>>> genode-main mailing list
>>>>>>> genode-main at lists.sourceforge.net
>>>>>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>>>>>
>>>>>>
>>>>>>
>>>>>> ------------------------------------------------------------------------------
>>>>>> _______________________________________________
>>>>>> genode-main mailing list
>>>>>> genode-main at lists.sourceforge.net
>>>>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>>>>
>>>>>
>>>>> ------------------------------------------------------------------------------
>>>>> _______________________________________________
>>>>> genode-main mailing list
>>>>> genode-main at lists.sourceforge.net
>>>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>>>
>>>>
>>>>
>>>> ------------------------------------------------------------------------------
>>>> _______________________________________________
>>>> genode-main mailing list
>>>> genode-main at lists.sourceforge.net
>>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>>
>>>
>>> ------------------------------------------------------------------------------
>>> _______________________________________________
>>> genode-main mailing list
>>> genode-main at lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>
>>
>>
>> ------------------------------------------------------------------------------
>> _______________________________________________
>> genode-main mailing list
>> genode-main at lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>
> 
> ------------------------------------------------------------------------------
> _______________________________________________
> genode-main mailing list
> genode-main at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/genode-main
> 





More information about the users mailing list