Philippe Gerum wrote:
> On Tue, 2010-08-24 at 18:46 +0200, Jan Kiszka wrote:
>> Philippe Gerum wrote:
>>> On Tue, 2010-08-24 at 18:30 +0200, Philippe Gerum wrote:
>>>> On Tue, 2010-08-24 at 17:59 +0200, Jan Kiszka wrote:
>>>>> Hi,
>>>>>
>>>>> just uploaded a forward port of the 2.6.34 ipipe patch for x86 to latest
>>>>> stable 2.6.35.3. It boots and runs fine here in 64-bit mode with Xenomai
>>>>> head, but I only ran light tests so far. Anyone interested in upgrading
>>>>> the host kernel (I think I read some request recently) is welcome to
>>>>> give it a try and report results back (specifically on 32 bit as that is
>>>>> a bit out of focus for me ATM). You can download the full git tree from
>>>>>
>>>>>   git://git.kiszka.org/ipipe-2.6.git queues/2.6.35-x86
>>>>>   (alternatively also via http://)
>>>>>
>>>>> Looking forward to feedback,
>>>> The comment and the relevant code for 82a7dd3df needs fixing: all
>>>> pipeline ports should expose 4 IPIs, named IPIPE_SERVICE_IPI[0-3].
>>>> powerpc/SMP has one more up to 2.6.34, but IPI4 will disappear in
>>>> 2.6.35. The upcoming arm/SMP port conforms to this requirement as well.
>>>> Those are merely pipelined IPI channels, the way the arch-dep section
>>>> manages to multiplex them (or not) over the available hw channel(s) is
>>>> of course unspecified. The virtual IPI numbers are also unspecified.
>>> Actually, the more I think of it, the less I see the value of checking
>>> the parameter passed to such an internal call like __ipipe_send_ipi().
>>> There is no user interaction with this code. So removing the test is
>>> indeed better.
>>>
>> Isn't ipipe_send_ipi() a public API? That's what I was thinking of: if
>> at all, then here.
> 
> Yes, ipipe_send_ipi() is the public API, calliong into
> __ipipe_send_ipi() for a per-arch implementation; I messed up in my
> explanation. My point is that your idea to remove the check from
> __ipipe_send_ipi seems correct to me, since nobody should send a silly
> value to this internal call. The test should be done in ipipe_send_ipi()
> once for all, relying on the generic names of the IPIs.
> 
> Regarding those names, and unlike I initially thought, ppc64 is still
> preventing me from removing IPI4 in 2.6.35, so I guess that for
> implementing a generic test, we would have to resort to #ifdef
> pollution.

Can we require that the IPI numbers are consecutive? Then the arch could
define an upper limit, and we could do a nicer range check in the
generic code.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

_______________________________________________
Adeos-main mailing list
[email protected]
https://mail.gna.org/listinfo/adeos-main

Reply via email to