CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

[Kyungmin Park]

[-- Attachment #1: Type: text/plain, Size: 533 bytes --]

Hi, Dirk Behme

As you mentioned, I tried to use CONFIG_PREEMPT_RT patch.
But there's some issues.

1. ethernet flags
  - After change from IRQF_RISING to IRQF_HIGH in recent OMAP2 patch. it occurs the GPIO bug message in line 1005.
    So I reverted IRQF_RISING as before. then it's working.

2. Too many loads???
  - After booting. I think the system is too slow. and also it's unstable. After some opeartions. it's hang.

3. ARMv6 support
  - current cmpxchg support for under ARMv6 only.

Thank you,
Kyungmin Park

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

[Dirk Behme]

Kyungmin Park wrote:
> As you mentioned, I tried to use CONFIG_PREEMPT_RT patch.

Great!

> But there's some issues.

Yep, CONFIG_PREEMPT_RT still looks experimental on our 
platform. Hopefully we contribute to improvement by testing it.

> 1. ethernet flags
>   - After change from IRQF_RISING to IRQF_HIGH in recent OMAP2 patch. it occurs the GPIO bug message in line 1005.
>     So I reverted IRQF_RISING as before. then it's working.

Maybe David has an idea here? [1]

Regarding ethernet I'm unsure. On OSK/OMAP1 my testcase is 
playing MP3 over NFS. While writing my previous mail, I 
never reached the end of MP3 because of NFS timeouts and 
connection losses.

Today, I tried again to get the exact error message and now 
no NFS issues any more. MP3 properly finishes.

> 2. Too many loads???
>   - After booting. I think the system is too slow. and also it's unstable. After some opeartions. it's hang.

Can't say this for OSK, feeling is same speed or even 
faster. For me, ALSA was unstable because I missed to 
convert some spinlocks to raw_spinlocks. Maybe you should 
check if you found all?

What's your testcase? For which operations does it hang?

> 3. ARMv6 support
>   - current cmpxchg support for under ARMv6 only.

Don't know anything about ARMv6.

Cheers

Dirk

[1] 
http://linux.omap.com/pipermail/linux-omap-open-source/2006-November/008270.html

Re: CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

[David Brownell]

On Monday 27 November 2006 11:09 am, Dirk Behme wrote:

> > 1. ethernet flags
> >   - After change from IRQF_RISING to IRQF_HIGH in recent OMAP2 patch. it occurs the GPIO bug message in line 1005.
> >     So I reverted IRQF_RISING as before. then it's working.
> 
> Maybe David has an idea here? [1]

No.  H4 with NFSroot and current GIT works fine for me; that
uses IRQF_TRIGGER_LOW.

The "best" solution would of course be passing the flags in
the IRQ resource, but that'd be more invasive to smc91x and
(eventually) the board init code.  platform_get_irq() doesn't
pass those flags along.

- Dave

Re: CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

[Dirk Behme]

Kyungmin Park wrote:
> 2. Too many loads???
>   - After booting. I think the system is too slow. and also it's unstable. After some opeartions. it's hang.

Two additional things:

- Do you converted OMAP2 timer to clocksource as well 
(timer-gp.c)? I think our clocksource patch only covers 
OMAP1 (Kevin?).

- Do yout get Cyclictest [1] to work?

Dirk

[1] http://rt.wiki.kernel.org/index.php/Cyclictest

Re: CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

[Kevin Hilman]

Dirk Behme wrote:
> Kyungmin Park wrote:
>> 2. Too many loads???
>>   - After booting. I think the system is too slow. and also it's
>> unstable. After some opeartions. it's hang.
> 
> Two additional things:
> 
> - Do you converted OMAP2 timer to clocksource as well (timer-gp.c)? I
> think our clocksource patch only covers OMAP1 (Kevin?).

Correct, only OMAP1 clocksource/clockevent done so far.

Kevin

Re: CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

[Kyungmin Park]

[-- Attachment #1: Type: text/plain, Size: 1088 bytes --]

Actually I used the 32K timer.

I tested it with fsstress on JFFS2 filesystem.


diff --git a/arch/arm/plat-omap/Kconfig b/arch/arm/plat-omap/Kconfig
index f0ce772..5cccb0c 100644
--- a/arch/arm/plat-omap/Kconfig
+++ b/arch/arm/plat-omap/Kconfig
@@ -11,9 +11,11 @@ choice

 config ARCH_OMAP1
        bool "TI OMAP1"
+       select GENERIC_TIME

 config ARCH_OMAP2
        bool "TI OMAP2"
+       select GENERIC_TIME if OMAP_32K_TIMER

 endchoice

Thank you,
Kyungmin Park



------- Original Message -------
Sender : Kevin Hilman<khilman@mvista.com> 
Date   : 2006-11-28 16:08
Title  : Re: CONFIG_PREEMPT_RT patch on Apollon (OMAP2)

Dirk Behme wrote:
> Kyungmin Park wrote:
>> 2. Too many loads???
>>   - After booting. I think the system is too slow. and also it's
>> unstable. After some opeartions. it's hang.
> 
> Two additional things:
> 
> - Do you converted OMAP2 timer to clocksource as well (timer-gp.c)? I
> think our clocksource patch only covers OMAP1 (Kevin?).

Correct, only OMAP1 clocksource/clockevent done so far.

Kevin



[-- Attachment #2: Type: text/plain, Size: 0 bytes --]