[Xenomai] Trying to get Xenomai on P1010RDB

[Xenomai] Trying to get Xenomai on P1010RDB

[Charles Retailleau]

Hi all,
The patching/compilation/installation steps seem fine.
But at the end I cannot have any Xenomai software running on the
P1010RDB (I tried most of the provided test SW).

I focused on latency and tried to debug it. I get that:

Program received signal SIGILL, Illegal instruction.
0x0ffde120 in __xn_rdtsc () from ....xenomai/lib/libnative.so.3

It appears that the 'mftb/mftbu' instructions
are in cause (cf. the small test in attachment, the 1st part works but
not the mftb test).
Furthermore when I check the core e500 ref manual
http://cache.freescale.com/files/32bit/doc/ref_manual/E500CORERM.pdf
MFTB appears has NOT being part of the instruction list.

Yet, Xenomai seems to work on P1020RDB which has the same core, so I'm
confused...
What did I do wrong ?

I'm using
   Xenomai 2.6.1,
   linux 3.2.21 configured with mpc85xx_defconfig (default options for
the rest of it),
   Freescale's SDK 1.1.2 (that's the main difference between me and
Lukasz Zemla's config on P1020RDB, so I'm trying to get the latest 1.2)

Thanks for your help.
Charles,
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: testMFTB.c
URL: <http://www.xenomai.org/pipermail/xenomai/attachments/20120816/d22f48c2/attachment.c>

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Lennart Sorensen]

On Thu, Aug 16, 2012 at 04:30:39PM +0200, Charles Retailleau wrote:
> Hi all,
> The patching/compilation/installation steps seem fine.
> But at the end I cannot have any Xenomai software running on the
> P1010RDB (I tried most of the provided test SW).
> 
> I focused on latency and tried to debug it. I get that:
> 
> Program received signal SIGILL, Illegal instruction.
> 0x0ffde120 in __xn_rdtsc () from ....xenomai/lib/libnative.so.3
> 
> It appears that the 'mftb/mftbu' instructions
> are in cause (cf. the small test in attachment, the 1st part works but
> not the mftb test).
> Furthermore when I check the core e500 ref manual
> http://cache.freescale.com/files/32bit/doc/ref_manual/E500CORERM.pdf
> MFTB appears has NOT being part of the instruction list.
> 
> Yet, Xenomai seems to work on P1020RDB which has the same core, so I'm
> confused...
> What did I do wrong ?
> 
> I'm using
>    Xenomai 2.6.1,
>    linux 3.2.21 configured with mpc85xx_defconfig (default options for
> the rest of it),
>    Freescale's SDK 1.1.2 (that's the main difference between me and
> Lukasz Zemla's config on P1020RDB, so I'm trying to get the latest 1.2)

mftb is a phased out instruction according to the powerpc 2.06 rev
B documentation.  mfspr is the prefered instruction to use instead on
everything other than the old 603 and POWER3 CPUs.

It is quite possible the e500 hence doesn't support it.  According to
http://cache.freescale.com/files/32bit/doc/app_note/AN3445.pdf it
specificly does NOT have mftb.  Neither does the e300 core which is what
we are using with xenomai (MPC8360e specificly).

-- 
Len Sorensen

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Wolfgang Grandegger]

On 08/16/2012 04:46 PM, Lennart Sorensen wrote:
> On Thu, Aug 16, 2012 at 04:30:39PM +0200, Charles Retailleau wrote:
>> Hi all,
>> The patching/compilation/installation steps seem fine.
>> But at the end I cannot have any Xenomai software running on the
>> P1010RDB (I tried most of the provided test SW).
>>
>> I focused on latency and tried to debug it. I get that:
>>
>> Program received signal SIGILL, Illegal instruction.
>> 0x0ffde120 in __xn_rdtsc () from ....xenomai/lib/libnative.so.3
>>
>> It appears that the 'mftb/mftbu' instructions
>> are in cause (cf. the small test in attachment, the 1st part works but
>> not the mftb test).
>> Furthermore when I check the core e500 ref manual
>> http://cache.freescale.com/files/32bit/doc/ref_manual/E500CORERM.pdf
>> MFTB appears has NOT being part of the instruction list.
>>
>> Yet, Xenomai seems to work on P1020RDB which has the same core, so I'm
>> confused...
>> What did I do wrong ?
>>
>> I'm using
>>    Xenomai 2.6.1,
>>    linux 3.2.21 configured with mpc85xx_defconfig (default options for
>> the rest of it),
>>    Freescale's SDK 1.1.2 (that's the main difference between me and
>> Lukasz Zemla's config on P1020RDB, so I'm trying to get the latest 1.2)
> 
> mftb is a phased out instruction according to the powerpc 2.06 rev
> B documentation.  mfspr is the prefered instruction to use instead on
> everything other than the old 603 and POWER3 CPUs.
> 
> It is quite possible the e500 hence doesn't support it.  According to
> http://cache.freescale.com/files/32bit/doc/app_note/AN3445.pdf it
> specificly does NOT have mftb.  Neither does the e300 core which is what
> we are using with xenomai (MPC8360e specificly).

It uses an e500v2 core. I use the ELDK 5.2.1 "powerpc-e500v2" for the P2020.

Wolfgang.

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Charles Retailleau]

Thanks Lennart that was very useful, it now works.
Here is the little patch I made, yet I couldn't find a switch to force
it for core e500/e500xx only...


--- ./xenomai-2.6.1.orig/include/asm-powerpc/syscall.h	2011-11-09
13:21:36.000000000 +0100
+++ ./xenomai-2.6.1.mine/./include/asm-powerpc/syscall.h	2012-08-17
15:47:28.387435909 +0200
@@ -167,9 +167,9 @@
 	} u;
 	unsigned long __tbu;

-	__asm__ __volatile__("1: mftbu %0\n"
-			     "mftb %1\n"
-			     "mftbu %2\n"
+	__asm__ __volatile__("1: mfspr %0,269\n"
+			     "mfspr %1,268\n"
+			     "mfspr %2,269\n"
 			     "cmpw %2,%0\n"
 			     "bne- 1b\n":"=r"(u.v[0]),
 			     "=r"(u.v[1]), "=r"(__tbu));
Thanks to all for your help
Charles,

Le 16/08/2012 16:46, Lennart Sorensen a écrit :
> On Thu, Aug 16, 2012 at 04:30:39PM +0200, Charles Retailleau
> wrote:
>> Hi all, The patching/compilation/installation steps seem fine. 
>> But at the end I cannot have any Xenomai software running on the 
>> P1010RDB (I tried most of the provided test SW).
>> 
>> I focused on latency and tried to debug it. I get that:
>> 
>> Program received signal SIGILL, Illegal instruction. 0x0ffde120
>> in __xn_rdtsc () from ....xenomai/lib/libnative.so.3
>> 
>> It appears that the 'mftb/mftbu' instructions are in cause (cf.
>> the small test in attachment, the 1st part works but not the mftb
>> test). Furthermore when I check the core e500 ref manual 
>> http://cache.freescale.com/files/32bit/doc/ref_manual/E500CORERM.pdf
>>
>> 
MFTB appears has NOT being part of the instruction list.
>> 
>> Yet, Xenomai seems to work on P1020RDB which has the same core,
>> so I'm confused... What did I do wrong ?
>> 
>> I'm using Xenomai 2.6.1, linux 3.2.21 configured with
>> mpc85xx_defconfig (default options for the rest of it), 
>> Freescale's SDK 1.1.2 (that's the main difference between me and 
>> Lukasz Zemla's config on P1020RDB, so I'm trying to get the
>> latest 1.2)
> 
> mftb is a phased out instruction according to the powerpc 2.06 rev 
> B documentation.  mfspr is the prefered instruction to use instead
> on everything other than the old 603 and POWER3 CPUs.
> 
> It is quite possible the e500 hence doesn't support it.  According
> to http://cache.freescale.com/files/32bit/doc/app_note/AN3445.pdf
> it specificly does NOT have mftb.  Neither does the e300 core which
> is what we are using with xenomai (MPC8360e specificly).
> 

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Gilles Chanteperdrix]

On 08/17/2012 04:24 PM, Charles Retailleau wrote:
> Thanks Lennart that was very useful, it now works.
> Here is the little patch I made, yet I couldn't find a switch to force
> it for core e500/e500xx only...
> 
> 
> --- ./xenomai-2.6.1.orig/include/asm-powerpc/syscall.h	2011-11-09
> 13:21:36.000000000 +0100
> +++ ./xenomai-2.6.1.mine/./include/asm-powerpc/syscall.h	2012-08-17
> 15:47:28.387435909 +0200
> @@ -167,9 +167,9 @@
>  	} u;
>  	unsigned long __tbu;
> 
> -	__asm__ __volatile__("1: mftbu %0\n"
> -			     "mftb %1\n"
> -			     "mftbu %2\n"
> +	__asm__ __volatile__("1: mfspr %0,269\n"
> +			     "mfspr %1,268\n"
> +			     "mfspr %2,269\n"
>  			     "cmpw %2,%0\n"
>  			     "bne- 1b\n":"=r"(u.v[0]),
>  			     "=r"(u.v[1]), "=r"(__tbu));
> Thanks to all for your help
> Charles,

If I understand correctly, this patch breaks old powerpcs. So, do we
have a compilation constant allowing to know whether we are compiling
for an old or new powerpc?

-- 
					    Gilles.

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Philippe Gerum]

On 08/17/2012 04:41 PM, Gilles Chanteperdrix wrote:
> On 08/17/2012 04:24 PM, Charles Retailleau wrote:
>> Thanks Lennart that was very useful, it now works.
>> Here is the little patch I made, yet I couldn't find a switch to force
>> it for core e500/e500xx only...
>>
>>
>> --- ./xenomai-2.6.1.orig/include/asm-powerpc/syscall.h	2011-11-09
>> 13:21:36.000000000 +0100
>> +++ ./xenomai-2.6.1.mine/./include/asm-powerpc/syscall.h	2012-08-17
>> 15:47:28.387435909 +0200
>> @@ -167,9 +167,9 @@
>>   	} u;
>>   	unsigned long __tbu;
>>
>> -	__asm__ __volatile__("1: mftbu %0\n"
>> -			     "mftb %1\n"
>> -			     "mftbu %2\n"
>> +	__asm__ __volatile__("1: mfspr %0,269\n"
>> +			     "mfspr %1,268\n"
>> +			     "mfspr %2,269\n"
>>   			     "cmpw %2,%0\n"
>>   			     "bne- 1b\n":"=r"(u.v[0]),
>>   			     "=r"(u.v[1]), "=r"(__tbu));
>> Thanks to all for your help
>> Charles,
>
> If I understand correctly, this patch breaks old powerpcs. So, do we
> have a compilation constant allowing to know whether we are compiling
> for an old or new powerpc?
>

For the sake of completeness, we could condition mftb to CONFIG_POWER3, 
all others we care for (i.e. >= ISA 2.01) do support mfspr 268.

-- 
Philippe.

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Gilles Chanteperdrix]

On 08/17/2012 05:02 PM, Philippe Gerum wrote:
> On 08/17/2012 04:41 PM, Gilles Chanteperdrix wrote:
>> On 08/17/2012 04:24 PM, Charles Retailleau wrote:
>>> Thanks Lennart that was very useful, it now works.
>>> Here is the little patch I made, yet I couldn't find a switch to force
>>> it for core e500/e500xx only...
>>>
>>>
>>> --- ./xenomai-2.6.1.orig/include/asm-powerpc/syscall.h	2011-11-09
>>> 13:21:36.000000000 +0100
>>> +++ ./xenomai-2.6.1.mine/./include/asm-powerpc/syscall.h	2012-08-17
>>> 15:47:28.387435909 +0200
>>> @@ -167,9 +167,9 @@
>>>   	} u;
>>>   	unsigned long __tbu;
>>>
>>> -	__asm__ __volatile__("1: mftbu %0\n"
>>> -			     "mftb %1\n"
>>> -			     "mftbu %2\n"
>>> +	__asm__ __volatile__("1: mfspr %0,269\n"
>>> +			     "mfspr %1,268\n"
>>> +			     "mfspr %2,269\n"
>>>   			     "cmpw %2,%0\n"
>>>   			     "bne- 1b\n":"=r"(u.v[0]),
>>>   			     "=r"(u.v[1]), "=r"(__tbu));
>>> Thanks to all for your help
>>> Charles,
>>
>> If I understand correctly, this patch breaks old powerpcs. So, do we
>> have a compilation constant allowing to know whether we are compiling
>> for an old or new powerpc?
>>
> 
> For the sake of completeness, we could condition mftb to CONFIG_POWER3, 
> all others we care for (i.e. >= ISA 2.01) do support mfspr 268.
> 

is CONFIG_POWER3 available when configuring user-space support?

-- 
					    Gilles.

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Philippe Gerum]

On 08/17/2012 05:11 PM, Gilles Chanteperdrix wrote:
> On 08/17/2012 05:02 PM, Philippe Gerum wrote:
>> On 08/17/2012 04:41 PM, Gilles Chanteperdrix wrote:
>>> On 08/17/2012 04:24 PM, Charles Retailleau wrote:
>>>> Thanks Lennart that was very useful, it now works.
>>>> Here is the little patch I made, yet I couldn't find a switch to force
>>>> it for core e500/e500xx only...
>>>>
>>>>
>>>> --- ./xenomai-2.6.1.orig/include/asm-powerpc/syscall.h	2011-11-09
>>>> 13:21:36.000000000 +0100
>>>> +++ ./xenomai-2.6.1.mine/./include/asm-powerpc/syscall.h	2012-08-17
>>>> 15:47:28.387435909 +0200
>>>> @@ -167,9 +167,9 @@
>>>>    	} u;
>>>>    	unsigned long __tbu;
>>>>
>>>> -	__asm__ __volatile__("1: mftbu %0\n"
>>>> -			     "mftb %1\n"
>>>> -			     "mftbu %2\n"
>>>> +	__asm__ __volatile__("1: mfspr %0,269\n"
>>>> +			     "mfspr %1,268\n"
>>>> +			     "mfspr %2,269\n"
>>>>    			     "cmpw %2,%0\n"
>>>>    			     "bne- 1b\n":"=r"(u.v[0]),
>>>>    			     "=r"(u.v[1]), "=r"(__tbu));
>>>> Thanks to all for your help
>>>> Charles,
>>>
>>> If I understand correctly, this patch breaks old powerpcs. So, do we
>>> have a compilation constant allowing to know whether we are compiling
>>> for an old or new powerpc?
>>>
>>
>> For the sake of completeness, we could condition mftb to CONFIG_POWER3,
>> all others we care for (i.e. >= ISA 2.01) do support mfspr 268.
>>
>
> is CONFIG_POWER3 available when configuring user-space support?
>

No, but actually, I never ported Xenomai to power3 and will likely never 
do this. It's the early 64bit server architecture; we won't officially 
support any 64bit platform which does not comply with ISA 2.04 at least 
anyway. Best effort would be to depend on a gcc define built-in for 
mcpu=power3 if any.

-- 
Philippe.

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Gilles Chanteperdrix]

On 08/17/2012 05:37 PM, Philippe Gerum wrote:
> On 08/17/2012 05:11 PM, Gilles Chanteperdrix wrote:
>> On 08/17/2012 05:02 PM, Philippe Gerum wrote:
>>> On 08/17/2012 04:41 PM, Gilles Chanteperdrix wrote:
>>>> On 08/17/2012 04:24 PM, Charles Retailleau wrote:
>>>>> Thanks Lennart that was very useful, it now works.
>>>>> Here is the little patch I made, yet I couldn't find a switch to force
>>>>> it for core e500/e500xx only...
>>>>>
>>>>>
>>>>> --- ./xenomai-2.6.1.orig/include/asm-powerpc/syscall.h	2011-11-09
>>>>> 13:21:36.000000000 +0100
>>>>> +++ ./xenomai-2.6.1.mine/./include/asm-powerpc/syscall.h	2012-08-17
>>>>> 15:47:28.387435909 +0200
>>>>> @@ -167,9 +167,9 @@
>>>>>    	} u;
>>>>>    	unsigned long __tbu;
>>>>>
>>>>> -	__asm__ __volatile__("1: mftbu %0\n"
>>>>> -			     "mftb %1\n"
>>>>> -			     "mftbu %2\n"
>>>>> +	__asm__ __volatile__("1: mfspr %0,269\n"
>>>>> +			     "mfspr %1,268\n"
>>>>> +			     "mfspr %2,269\n"
>>>>>    			     "cmpw %2,%0\n"
>>>>>    			     "bne- 1b\n":"=r"(u.v[0]),
>>>>>    			     "=r"(u.v[1]), "=r"(__tbu));
>>>>> Thanks to all for your help
>>>>> Charles,
>>>>
>>>> If I understand correctly, this patch breaks old powerpcs. So, do we
>>>> have a compilation constant allowing to know whether we are compiling
>>>> for an old or new powerpc?
>>>>
>>>
>>> For the sake of completeness, we could condition mftb to CONFIG_POWER3,
>>> all others we care for (i.e. >= ISA 2.01) do support mfspr 268.
>>>
>>
>> is CONFIG_POWER3 available when configuring user-space support?
>>
> 
> No, but actually, I never ported Xenomai to power3 and will likely never 
> do this. It's the early 64bit server architecture; we won't officially 
> support any 64bit platform which does not comply with ISA 2.04 at least 
> anyway. Best effort would be to depend on a gcc define built-in for 
> mcpu=power3 if any.
> 

Ok. According to ppc64-linux-gcc from ELDK 4.2, all architectures above
power4 have _ARCH_PWR4 defined.

-- 
					    Gilles.

Re: [Xenomai] Trying to get Xenomai on P1010RDB

[Gilles Chanteperdrix]

On 08/17/2012 04:24 PM, Charles Retailleau wrote:

> Thanks Lennart that was very useful, it now works.
> Here is the little patch I made, yet I couldn't find a switch to force
> it for core e500/e500xx only...
> 
> 
> --- ./xenomai-2.6.1.orig/include/asm-powerpc/syscall.h	2011-11-09
> 13:21:36.000000000 +0100
> +++ ./xenomai-2.6.1.mine/./include/asm-powerpc/syscall.h	2012-08-17
> 15:47:28.387435909 +0200
> @@ -167,9 +167,9 @@
>  	} u;
>  	unsigned long __tbu;
> 
> -	__asm__ __volatile__("1: mftbu %0\n"
> -			     "mftb %1\n"
> -			     "mftbu %2\n"
> +	__asm__ __volatile__("1: mfspr %0,269\n"
> +			     "mfspr %1,268\n"
> +			     "mfspr %2,269\n"
>  			     "cmpw %2,%0\n"
>  			     "bne- 1b\n":"=r"(u.v[0]),
>  			     "=r"(u.v[1]), "=r"(__tbu));
> Thanks to all for your help
> Charles,


Merged. Thanks.

-- 
                                                                Gilles.