Re: [Xenomai-help] Moving from 2.6.14(ppc) to 2.6.20(powerpc): illicit calls detected by ipipe

[Philippe Gerum <rpm@xenomai.org>]

Johan Borkhuis wrote:
> Philippe Gerum wrote:
>> Johan Borkhuis wrote:
>>  
>>> Hello,
>>>
>>> I am trying to move from 2.6.14(ppc) to 2.6.20(powerpc). On ppc I am
>>> using Xenomai 2.4.1 and I-Pipe version 1.5-03, on powerpc I am using
>>> Xenomai 2.4.1, I-Pipe version 1.7-12.
>>> Most of the system is working, but I am having some problems with one
>>> device driver. This is a "standard" Linux driver, that I converted
>>> into a Xenomai driver. This driver does a very limited
>>> initialization, the main init is done when the device is opened and
>>> after some IOCTL calls.
>>>
>>> The problem is that I am getting "Detected illicit call from domain
>>> 'Xenomai' into a service reserved for domain 'Linux' and below."
>>> messages from Adeos, which I did not get when using 2.6.14. The code
>>> I use is identical, except for some small areas where the API kernel
>>> changed between 2.6.14 and 2.6.20. Has there been a change between
>>> the versions of Adeos or Xenomai which could cause this change in
>>> behavior?
>>> (BTW: this message is generated when the option IPIPE_DEBUG_CONTEXT
>>> is on, when this option is off I get a kernel Oops in
>>> kernel/ipipe/core.c:210.
>>>
>>> The affected calls are calls like kmalloc and kfree, but also calls
>>> like check_mem_region, which uses kmem_cache_zalloc inside the
>>> function. Normally these should not be used inside a RT application,
>>> but this is the startup phase of the application and there are no RT
>>> restraints at this point.
>>>
>>>     
>>
>> 2.6.14 did not provide any invalid call context detection, so the fact
>> that you
>> start getting those with 2.6.20 is just the sign that your driver had
>> a problem
>> since 2.6.14, but they went unnoticed so far.
>>   
> 
> They were unnoticed in 2.6.14, but do they cause any trouble? As far as
> I can see the driver did work correctly in 2.6.14.

Yes, this is why a detector was added. This kind of error is a source of random
kernel corruption.

> Can I just ignore this error in 2.6.20, and have the driver continue, or
> does this situation generate an error, so that my application cannot
> continue?

You should really fix it.

> 
>> You may want to send the stack backtrace dumped by the detector, so
>> that we
>> could clearly identify what the calling context is.
>>   
> 
> Below is the call-trace:
> 
> I-pipe: Detected illicit call from domain 'Xenomai'
>        into a service reserved for domain 'Linux' and below.
> Call Trace:
> [97A479E0] [8000817C] show_stack+0x7c/0x1a4 (unreliable)
> [97A47A10] [8005C2A4] ipipe_check_context+0xb8/0xc0
> [97A47A30] [800D88F0] kmem_cache_zalloc+0xe4/0xf4
> [97A47A50] [80033520] __request_region+0x48/0xd4
> [97A47A70] [800335E0] __check_region+0x34/0x6c
> [97A47A80] [A1216F94] men_OSS_AssignResources+0xec/0x1dc [men_oss]
> [97A47AB0] [A12309BC] A201_Init+0x618/0x708 [men_bb_a201]
> [97A47B10] [A12319F0] InitWrapper+0x38/0x50 [men_bb_a201]
> [97A47B20] [A122C5E4] OpenDevice+0x4a8/0x634 [men_bbis_kernel]
> [97A47C20] [A122CF9C] bbis_open+0x130/0x220 [men_bbis_kernel]
> [97A47C60] [A126B450] mk_ioctl_core+0x548/0xcec [men_mdis_kernel]
> [97A47DB0] [A126FA6C] mdisctl_ioctl_rt+0x84/0x1b40 [men_mdis_kernel]

If you want to call into the regular Linux services from a rtdm-based ioctl
routine, you should define your routine in the _nrt section not in the _rt one.
E.g.
struct rtdm_device device = {
	...
	.ops = {
		.ioctl_nrt =      mdisctl_ioctl,

If you really need part of your ioctl code to run in RT mode, then you should
split it in two separate routines, one declared in the _rt section, and the
other one in the _nrt section.

> [97A47DE0] [800A5D04] __rt_dev_ioctl+0xb0/0x1bc
> [97A47E90] [800A8A3C] sys_rtdm_ioctl+0x3c/0x54
> [97A47EA0] [80074338] losyscall_event+0xe4/0x208
> [97A47ED0] [8005D88C] __ipipe_dispatch_event+0x1ec/0x2a4
> [97A47F30] [80012268] __ipipe_syscall_root+0x5c/0x13c
> [97A47F40] [80010A5C] DoSyscall+0x24/0x60
> 
> And another one (I replaced the kfree with rtdm_free, and the associated
> kalloc with rtdm_alloc, after that this message disappeared and the
> message above appeared):
> I-pipe: Detected illicit call from domain 'Xenomai'
>        into a service reserved for domain 'Linux' and below.
> Call Trace:
> [9758BBD0] [8000732C] show_stack+0x50/0x188 (unreliable)
> [9758BC00] [8004ED48] ipipe_check_context+0x88/0x8c
> [9758BC10] [800B6224] kfree+0x24/0xec
> [9758BC30] [A10A99AC] bbis_open+0x11c/0x550 [men_bbis_kernel]

You are still calling kfree() at some point in the routine above.

> [9758BC60] [A10EB13C] mk_ioctl_core+0x52c/0xcd0 [men_mdis_kernel]
> [9758BDB0] [A10EF384] mdisctl_ioctl_rt+0x6c/0x17c0 [men_mdis_kernel]
> [9758BDE0] [8008B360] __rt_dev_ioctl+0x98/0x198
> [9758BE90] [8008D808] sys_rtdm_ioctl+0x20/0x30
> [9758BEA0] [80061230] losyscall_event+0xc8/0x1e4
> [9758BED0] [8004FDDC] __ipipe_dispatch_event+0x10c/0x1e4
> [9758BF30] [8000FAE4] __ipipe_syscall_root+0x40/0xe8
> [9758BF40] [8000E610] DoSyscall+0x24/0x60
> 
> 
>>  
>>> Kind regards,
>>>     Johan Borkhuis
>>>
>>>     
>>
>>
>>   
> 
> 


-- 
Philippe.