Re: [Xenomai] Porting Ipipe to new ARM SoC (Xilinx Zynq)

[Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>]

On 01/23/2013 04:43 AM, Matthew Fornero wrote:

> I'll cat out the proc/interrupts when I get to the office tomorrow. I
> did read the porting guide, but stupidly missed that function call. On
> this dev kit, we're not initially planning to use GPIO from our code
> (real time or otherwise), but of course they could be used by various
> peripherals.
> 
> I'm somewhat new to arm (mostly worked on x86 until now), so please
> excuse my ignorance-- would there be another irq chip or something I
> need to multiplex in other driver files (other than the
> gpio-xilinxps.c file)? Is there an existing platform I can reference
> for this? Maybe some code in the board level file as well?


Each platform defines its own irqchips for multiplexed GPIOs. If you
want to track all invalid GPIO demuxers, you can enable ipipe debugging
and add ipipe_root_only() inside "generic_handle_irq".

> 
> I'll try the remainder of the above debug steps as well. Unfortunately
> our JTAG environment isn't totally stable, but I should be able to get
> something going.
> 
>>> If I boot with Xenomai off (CONFIG_XENOMAI = n) but IPIPE still on
>>> (CONFIG_IPIPE = y), I get a warning at boot. Haven't looked into the
>>> root cause of this-- but maybe it's related?
>>>
>>> mmc0: new high speed SDHC card at address b368
>>> mmcblk0: mmc0:b368 F0F0F 3.71 GiB
>>>  mmcblk0: p1 p2
>>> EXT4-fs (mmcblk0p2): recovery complete
>>> EXT4-fs (mmcblk0p2): mounted filesystem with ordered data mode. Opts:
>>> (null)
>>> VFS: Mounted root (ext4 filesystem) on device 179:2.
>>> devtmpfs: mounted
>>> Freeing init memory: 160K
>>> ------------[ cut here ]------------
>>> WARNING: at arch/arm/mm/context.c:182 __new_context+0x40/0x118()
>>
>>
>> Known issue, which I thought was fixed, but maybe not. Try this patch:
>>
>> diff --git a/arch/arm/mm/context.c b/arch/arm/mm/context.c
>> index 3ae015f..79de5dc 100644
>> --- a/arch/arm/mm/context.c
>> +++ b/arch/arm/mm/context.c
>> @@ -167,11 +167,12 @@ static inline void set_mm_context(struct mm_struct
>> *mm, unsigned int asid)
>>
>>  void __new_context(struct mm_struct *mm)
>>  {
>> -       int cpu = ipipe_processor_id();
>>         unsigned long flags;
>>         unsigned int asid;
>> +       int cpu;
>>
>>         asid_lock(flags);
>> +       cpu = ipipe_processor_id();
>>  #ifdef CONFIG_SMP
>>         /*
>>          * Check the ASID again, in case the change was broadcast from
> 
> Will try tomorrow and report if it resolves it.


Forget it, it is purely a problem due to CONFIG_XENOMAI=n and
CONFIG_IPIPE=y, in that case preemptible context switches do not get
selected, __new_context gets called with irqs off, which can not work
with CONFIG_IPIPE. The fix is to select IPIPE_WANT_PREEMPTIBLE_SWITCH
with CONFIG_SMP, I thought I did it, but it probably got lost.

> 
>>
>>> If I boot with Xenomai on (CONFIG_XENOMAI = y) and Nucleus as a module
>>> (CONFIG_XENO_OPT_NUCLEUS = m), the system boots without issue (not even
>>> a warning at boot), but the Nucleus module fails to compile with the
>>> following error:
>>>
>>> [mfornero@hwbuild linux]$ make -j8 modules
>>>   CHK     include/linux/version.h
>>>   CHK     include/generated/utsrelease.h
>>> make[1]: `include/generated/mach-types.h' is up to date.
>>>   CALL    scripts/checksyscalls.sh
>>>   Building modules, stage 2.
>>>   MODPOST 22 modules
>>> ERROR: "current_mm" [kernel/xenomai/nucleus/xeno_nucleus.ko] undefined!
>>> make[1]: *** [__modpost] Error 1
>>> make: *** [modules] Error 2
>>
>>
>> It simply means EXPORT_SYMBOL_GPL(current_mm) is missing.
> 
> So this export should go in something like arch/arm/include/asm/mmu_context.h?
> 
> Do I need to do anything special for the export given that it's
> defined via "DECLARE_PER_CPU"?


Right, you need to use EXPORT_PER_CPU_SYMBOL_GPL(current_mm).

-- 
                                                                Gilles.