set pager in foc

Stefan Kalkowski stefan.kalkowski at ...1...
Fri Jul 1 12:21:27 CEST 2011 

On 01.07.2011 01:51, Chen Tian wrote:
> Hi Stefan,
> 
> Thank you for the reply. I think I have understood most of it. What is still
> confusing to me is "badge". Apparently there are two different badges. One
> is defined in Capability node, which is essentially the label of the IPC
> gate of a Pager object. It is also stored in
> Pager_object::cap()::local_name() and is used by the pager thread to look up
> the Pager_object from the pager entrypoint. I wonder if there is any
> particular reason why the badge values cannot be consecutive numbers
> (BADGE_MASK = ~3)? Also, why does the label returned from IPC call need to
> perform " & ~3UL"? After reading the L4RE document, I don't think kernel
> will make any change to this value. 

Hi Chen,

the Fiasco.OC kernel internally uses the two least significant bits of a
IPC-gate's label as flags. I don't know anymore what exactly was encoded
into these, somehow the rights of the capability, but we do not
interpret them in Genode (probably L4Re does). Nevertheless, these two
bits might get overridden, so we cannot rely on them.

> 
> The second badge is the one used in the Pager_object. It is defined as a cap
> index selector of the native foc thread (rm_session_component.cc), but I
> cannot see how this one is used in foc.  
> 

The badge of a Pager_object is used in the Fiasco.OC specific code, to
store the (Fiasco.OC-)capability referencing the failing thread.
Normally, as a pager you simply reply to a pagefault, and thereby use
the implicit, hidden reply-capability. But in Genode sometimes we have
to solve pagefaults asynchronously, thereby loosing the implicit
reply-capability. That's why we use the badge member of the generic
Pager_object implementation in the Fiasco.OC specific case as to store a
capability referencing the pager's client.

I hope that clarifies your question.

Regards
Stefan

> Best,
> Chen
> 
> 
> 
> -----Original Message-----
> From: Stefan Kalkowski [mailto:stefan.kalkowski at ...1...] 
> Sent: Thursday, June 30, 2011 4:29 AM
> To: genode-main at lists.sourceforge.net
> Subject: Re: set pager in foc
> 
> On 29.06.2011 23:00, Chen Tian wrote:
>> Hi Stefan,
>>
>> So that means each Pager_object also has a corresponding Fiasco OC kernel
>> object, which can do the IPC to the pager thread, right?
> 
> Hi Chen,
> 
> each Pager_object has an own IPC-gate (kernel-object) to the
> pager-thread, the pager-thread thereby distinguishes different faulters
> from each other.
> 
>> By looking at the
>> cap_session_component.cc, I think the kernel object is essentially an
>> IPC-gate to the pager thread, i.e., _pager_ep->activation. Can you briefly
>> tell me how the ipc-gate works? I think I will spend some time on the
> kernel
>> later. Thanks a lot.
> 
> An IPC-gate is some kind of a thread's wrapper object. In addition to
> the target thread object, you can assign a label to it, during its
> creation. The label is received by the destination thread whenever IPC
> is send through the IPC-gate. This label value cannot be altered by some
> (remote) owner of the IPC-capability. Moreover, when owning the IPC-gate
> you cannot e.g. destroy the corresponding thread (which you can do, if
> you own the "real" thread capability), but only do IPC with it.
> 
> Regards
> Stefan
> 
>>
>> Best,
>> Chen
>>
>>
>>
>> -----Original Message-----
>> From: Stefan Kalkowski [mailto:stefan.kalkowski at ...1...] 
>> Sent: Tuesday, June 28, 2011 2:22 AM
>> To: Genode OS Framework Mailing List
>> Subject: Re: set pager in foc
>>
>> Hi Chen,
>>
>> On 27.06.2011 23:17, Chen Tian wrote:
>>> Hi Stefan,
>>>
>>> Sorry for the late response, as I am busy with something else.
>>>
>>> Can you please explain to me what _pager->cap().dst() is and when it is
>>> defined? 
>>
>> '_pager' is a 'Pager_object' - an object which is local to the pager,
>> and encapsulates the state of a pager's client. It is referenced from
>> outside (e.g. by a failing client) via the pager-capability.
>> The 'Pager_object' has a 'Genode::Capability' member, which is returned
>> by 'Pager_object::cap()'. This is the capability which references the
>> 'Pager_object'. The 'Genode::Capability' is Genode's own abstraction of
>> a capability, and shouldn't be mixed up with the capability provided by
>> the Fiasco.OC kernel. Nevertheless, every Genode capability internally
>> uses a kernel capability. The kernel capability can be accessed via
>> 'Genode::Capability::dst()'.
>>
>> So in short _pager->cap().dst() gives you the capability, that
>> references the pager-thread kernel-object, which can be used e.g. to do
>> IPC to that thread. The 'Genode::Capability' member of the Pager_object
>> is set, when it gets associated with a pager, have a look at the
>> Pager_entrypoint::manage(...) function in:
>>
>>   base-foc/src/base/pager/pager.cc
>>
>> In core this is done, whenever someone calls
>> 'add_client(Thread_capability thread)' via the rm_session interface.
>> Have a look at here:
>>
>>   base/src/core/rm_session_component.cc
>>
>> the essential line is the following:
>>
>>   return Pager_capability(_pager_ep->manage(cl));
>>
>>> Is this the global pager thread we are talking about?
>>
>> If you will, yes. (Once again we've to restrict this statement to core's
>> pager implementation on Fiasco.OC. ;-) )
>>
>> I hope it helped you to better understand the code around the paging
> issues.
>>
>> Regards
>> Stefan
>>
>>> I am looking
>>> at add_client function and found that only Pager_object::_badge is
>>> initialized through Rm_client constructor. Thanks.
>>>
>>> Best,
>>> Chen
>>>
>>>
>>> -----Original Message-----
>>> From: Stefan Kalkowski [mailto:stefan.kalkowski at ...1...] 
>>> Sent: Wednesday, June 22, 2011 2:09 AM
>>> To: Genode OS Framework Mailing List
>>> Subject: Re: set pager in foc
>>>
>>> Hi Chen,
>>>
>>> On 22.06.2011 02:54, Chen Tian wrote:
>>>> Thanks for the reply, two more questions. 
>>>>
>>>> First, why _pager is considered global? Isn't it constructed for each
>>> thread
>>>> as a client of an RM session?
>>>
>>> Yeah you're right. That's why I put global pager in double quotes. With
>>> global pager I meant that for almost all applications in
>>> Genode/Fiasco.OC there is only one pager thread in core, like you've
>>> stated before. So the local member _pager is mostly the same capability
>>> for all different thread objects (except for instance L4Linux).
>>>
>>>> Second, the pager thread whose entry() is defined in pager.cc should be
>>> the
>>>> one that receives the page fault request from the kernel. But how is
> this
>>>> thread registered into kernel so that the kernel knows its existence?
>>>
>>> At first, the pager thread is a normal thread which starts up and waits
>>> for IPC messages. In Fiasco.OC one can send IPC messages to a thread
>>> only, when owning a capability for that thread. Whenever a thread raises
>>> a pagefault, the kernel delivers an IPC-message to the pager of that
>>> thread. Therefore the kernel uses a capability, that was registered as
>>> the thread's pager thread beforehand. This registration of the pager
>>> capability is done via the l4_thread_control_* system call.
>>>
>>> Does that clarify your question?
>>>
>>> regards Stefan
>>>
>>>>
>>>> Thanks for the help.
>>>>
>>>> Best,
>>>> Chen
>>>>
>>>> -----Original Message-----
>>>> From: Stefan Kalkowski [mailto:stefan.kalkowski at ...1...] 
>>>> Sent: Monday, June 20, 2011 7:10 AM
>>>> To: Genode OS Framework Mailing List
>>>> Subject: Re: set pager in foc
>>>>
>>>> Hello Chian,
>>>>
>>>> On 18.06.2011 01:38, Chen Tian - SISA wrote:
>>>>> Hello,
>>>>>
>>>>>  
>>>>>
>>>>> After the set_pager function of a thread is called through a
>>>>> cpu_session, the _pager  of that thread is set and used when the actual
>>>>> platform thread starts. 
>>>>
>>>> That's true at least for the implementation of the rm_session interface
>>>> in core with respect to Genode/Fiasco.OC.
>>>>
>>>>> But there is only one thread (pager) to resolve
>>>>> all page faults for all threads.
>>>>
>>>> This is almost true for the Fiasco.OC version of Genode. Like Norman
>>>> explained in a previous mail, other platforms don't necessarily have
>>>> only one pager. In the current implementation for Fiasco.OC this is true
>>>> for normal Genode applications. L4Linux for example pages its VCPUs, and
>>>> applications itself. A VCPU (some kind of special thread concept in
>>>> recent Fiasco.OC revisions) is constructed by Genode's core process, but
>>>> afterwards the capability slot containing the global pager capability
>>>> gets overmapped to contain a special pager thread, which is running in
>>>> the L4Linux-kernel protection-domain.
>>>>
>>>>> I think the pager mapping  happened in
>>>>> platform_thread::start (platform_thread.cc:40-47). Can somebody please
>>>>> explain to me the meaning of this l4_task_map call?
>>>>
>>>> In Fiasco.OC a pager gets assigned to a thread by choosing a capability
>>>> slot in the capability space of the target protection domain, wherein
>>>> the new thread will run. The capability in that slot should point to the
>>>> pager thread. That why we first map the capability of the pager (which
>>>> was set by the cpu_session interface first) into the capability space of
>>>> the target protection domain. This is what the following snipet does:
>>>>
>>>>   l4_task_map(_platform_pd->native_task(), L4_BASE_TASK_CAP,
>>>>               l4_obj_fpage(_pager->cap().dst(), 0, L4_FPAGE_RWX),
>>>>               _remote_pager_cap | L4_ITEM_MAP);
>>>>
>>>> After that, you instruct the kernel to use the capability slot, we just
>>>> filled with life, as pager and exception-handler:
>>>>
>>>>   l4_thread_control_start();
>>>>   l4_thread_control_pager(_remote_pager_cap);
>>>>   l4_thread_control_exc_handler(_remote_pager_cap);
>>>>   l4_thread_control_bind(_utcb, _platform_pd->native_task());
>>>>
>>>>> I suspect that
>>>>> “_remote_pager_cap” is corresponding to  the global pager, but cannot
>>>>> find any evidence.  Thanks a lot.
>>>>
>>>> The '_remote_pager_cap' denotes the capability slot in the thread's
>>>> capability space, whereby '_pager' is the capability slot in core's
>>>> capability space, that gets mapped to '_remote_pager_cap' by
>>>> l4_task_map. The '_pager' variable is set via the cpu_session interface,
>>>> like you've stated before. Although, this is normally the "one global
>>>> pager", Genode's design and the Fiasco.OC specific implementation is not
>>>> limited to use only that one pager thread.
>>>>
>>>> Best regards
>>>> Stefan
>>>>
>>>>>
>>>>>  
>>>>>
>>>>> Best,
>>>>>
>>>>> Chen
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>
>>
> ----------------------------------------------------------------------------
>>>> --
>>>>> EditLive Enterprise is the world's most technically advanced content
>>>>> authoring tool. Experience the power of Track Changes, Inline Image
>>>>> Editing and ensure content is compliant with Accessibility Checking.
>>>>> http://p.sf.net/sfu/ephox-dev2dev
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Genode-main mailing list
>>>>> Genode-main at lists.sourceforge.net
>>>>> https://lists.sourceforge.net/lists/listinfo/genode-main
>>>>
>>>
>>
> 

-- 
Stefan Kalkowski
Genode Labs

http://www.genode-labs.com/ · http://genode.org/

Genode Labs GmbH · Amtsgericht Dresden · HRB 28424 · Sitz Dresden
Geschäftsführer: Dr.-Ing. Norman Feske, Christian Helmuth

More information about the users mailing list