[Adeos-main] RE: RTAI/fusion 0.7

[Adeos-main] RE: RTAI/fusion 0.7

[Fillod Stephane]

[-- Attachment #1: Type: text/plain, Size: 654 bytes --]

Hi Philippe,

Kudos for the RTAI/fusion 0.7 release!
You (the team) are doing an astounding work.

[snip]
>o The Adeos support used in fusion has been upgraded to final releases,

>namely r9/x86 and r6/ppc. Upgrading is strongly recommended, especially

>if CONFIG_PREEMPT is enabled in your kernel config, which is likely
with 
>fusion. Also note that Adeos PPC is now rebased on mainline (2.6.10 
>actually).

Please find attached a patch of Adeos r6/ppc against 2.6.11.
I hope I've got it right. Please let me know if it would need 
the spinlock fix in i386-r10c1.

Hope to get some Fusion report to share soon.

-- 
Stephane


[-- Attachment #2: adeos-linux-2.6.11-ppc-r6-sf.patch.gz --]
[-- Type: application/x-gzip, Size: 38593 bytes --]

Re: [Adeos-main] RE: RTAI/fusion 0.7

[Heikki Lindholm]

Fillod Stephane kirjoitti:
> Hi Philippe,
> 
> Kudos for the RTAI/fusion 0.7 release!
> You (the team) are doing an astounding work.
> 
> [snip]
> 
>>o The Adeos support used in fusion has been upgraded to final releases,
> 
> 
>>namely r9/x86 and r6/ppc. Upgrading is strongly recommended, especially
> 
> 
>>if CONFIG_PREEMPT is enabled in your kernel config, which is likely
> 
> with 
> 
>>fusion. Also note that Adeos PPC is now rebased on mainline (2.6.10 
>>actually).
> 
> 
> Please find attached a patch of Adeos r6/ppc against 2.6.11.
> I hope I've got it right. Please let me know if it would need 
> the spinlock fix in i386-r10c1.

The patch seemed ok, but the kernel isn't. There's a bug that causes the 
plain kernel schedule while in atomic region (see
http://ozlabs.org/pipermail/linuxppc-dev/2005-March/018497.html
)
I don't know if this affects embedded archs, but anything using entry.S 
seems to be affected. I wouldn't use 2.6.11 on ppc32.

-- Heikki Lindholm

Re: [Adeos-main] RE: RTAI/fusion 0.7

[Philippe Gerum]

Hi Stéphane,

Fillod Stephane wrote:
> Hi Philippe,
> 
> Kudos for the RTAI/fusion 0.7 release!
> You (the team) are doing an astounding work.

Thanks.

> 
> [snip]
> 
>>o The Adeos support used in fusion has been upgraded to final releases,
> 
> 
>>namely r9/x86 and r6/ppc. Upgrading is strongly recommended, especially
> 
> 
>>if CONFIG_PREEMPT is enabled in your kernel config, which is likely
> 
> with 
> 
>>fusion. Also note that Adeos PPC is now rebased on mainline (2.6.10 
>>actually).
> 
> 
> Please find attached a patch of Adeos r6/ppc against 2.6.11.
> I hope I've got it right.

Looks so. Looking at the problem Heikki pointed us at, it would somewhat 
affect 2.6.11 usability in RTAI/fusion context, since we would need to 
run with CONFIG_PREEMPT disabled. No problem for RT stuff in primary 
mode, but rather hopeless for using the secondary one efficiently. 
Either we include Paul Mackerras's fix for it in the Adeos patch, or we 
go directly for 2.6.12-rc, using your patch as a reference.

> Please let me know if it would need 
> the spinlock fix in i386-r10c1.
> 

It would be better, yes. This generic bug breaks deemed non-preemptible 
sections in printk() at least. I've committed the changes to the Adeos 
CVS (adeos-core/include/asm-ppc/adeos.h).

> Hope to get some Fusion report to share soon.
> 

Ack. Thanks for the patch.

-- 

Philippe.