Re: [Xenomai] Upgrading to Xenomai 2.6.1

[Doug Brunner <dbrunner@ebus.com>]

On 09/04/2012 01:09 AM, Gilles Chanteperdrix wrote:
> On 09/04/2012 09:27 AM, Doug Brunner wrote:
>> On 09/02/2012 10:28 PM, Gilles Chanteperdrix wrote:
>>> On 09/03/2012 01:48 AM, Doug Brunner wrote:
>>>
>>>> On 08/31/2012 12:39 AM, Gilles Chanteperdrix wrote:
>>>>> On 08/31/2012 08:40 AM, Doug Brunner wrote:
>>>>>
>>>>>> I've been having a very hard time with the upgrade from Xenomai 2.6.0 to
>>>>>> 2.6.1--hoping someone might have a suggestion for me. My thinking was
>>>>>> that, to get the most possible bug fixes, I should also upgrade to the
>>>>>> latest kernel for which there is a stable I-pipe patch, which appears to
>>>>>> be 3.2.21 from examining the list of available patches distributed with
>>>>>> Xenomai.
>>>>>>
>>>>>> Unfortunately, 3.2.21 (even unmodified, using a vanilla Debian kernel
>>>>>> config, set for 586 cpu type) will not boot on my target platform
>>>>>> (Winsystems PPM-LX800-G, which has a Geode LX800 CPU). It immediately
>>>>>> resets after "Loading initial ramdisk", without even any early printk
>>>>>> output. I have no idea what could be causing that :(
>>>>> Ok. I have just booted my geode with 3.4, both with 586 cpu type without
>>>>> tsc and with geode cpu type with tsc. The kernel boots and the latency
>>>>> test starts. So, I guess you selected a kernel configuration option we
>>>>> do not support (such as CONFIG_KGDB).
>>>>>
>>>>> You can download the configuration which works for me here:
>>>>> http://xenomai.org/~gch/config-geode
>>>>>
>>>> Argh...the problem was apparently that I'd left SMP support selected.
>>>> Thanks for your config! It's a nice bare-bones kernel and boots my Geode
>>>> as well so I'm using it as a new starting point.
>>>> I saw an email from you on 8/3 referring to a regression connected to
>>>> irq_hold that was fixed in your git. Should I get an I-pipe patch from
>>>> there, or from git.denx.de/ipipe.git instead of using the one
>>>> distributed with 2.6.1?
>>> Probably not, that is a fix for IO-APIC interrupt controller, the geode
>>> does not have an IO-APIC. Making a few tests with geode without tsc I
>>> found out something obvious: "statistics collection" should be disabled
>>> as it causes the tsc read function to be called often, and each read of
>>> the emulated tsc costs 4us.
>> If I understand you correctly you are recommending I use 3.2.21 with the
>> ipipe patch supplied with Xeno 2.6.1 (changing ipipe_root_domain_p to
>> ipipe_root_p in arch/x86/lib/mmx_32.c).
>>
>> Unfortunately it seems neither that 3.2.21 nor the already ipipe-enabled
>> 3.4.6 from git.denx.de/ipipe.git
>> (29e656a9501663366287a893d5f6d62bff9f40f6) works with my target
>> Winsystems PPM-LX800-G. Whether I use the config you posted at
>> http://xenomai.org/~gch/config-geode or mine (derived from yours, just
>> drops some ethernet devices I don't use and adds the Geode framebuffer
>> and IPv6), the system experiences ATA faults such as the following
>> immediately upon trying to boot:
>>
>> ata1.00: exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen
>> ata1.00: failed command: READ DMA
>> ata1.00: cmd c8/00:08:00:00:00/00:00:00:00:00/e0 tag 0 dma 4096 in
>>            res 40/00:00:00:00:00/00:00:00:00:00/00 Emask 0x4 (timeout)
>> ata1.00: status: { DRDY }
>>
>> Finally, after several of these with long delays in between, it panics
>> because it can't find the root filesystem on the ATA device it's trying
>> to access (CF card via the CS5536 controller).
>>
>> The same config minus the ipipe stuff, fed to an unpatched kernel tree,
>> produces a binary that will boot the system with no error messages; this
>> is true for both 3.2.21 and 3.4.6. Any idea what could be doing this, or
>> what else I can do to debug? I've attached my configs for reference,
>> except the 3.4.6 ipipe which I forgot to save.
> The cs5536 SATA driver does not work for me either, even without I-pipe,
> whereas the legacy IDE driver (option CONFIG_BLK_DEV_CS5536) works,
> could you test whether it works for you?
>
For me the CS5536 PATA driver (not SATA as you have) works fine without 
I-pipe, but with I-pipe I get the fault described above. The legacy IDE 
driver does work with I-pipe, at least on 3.2.21. I will be running the 
Xenomai test suite against that build tomorrow.