In article <46AA135C.1000505@cat.co.za>, Ian McIntosh  <ianm@cat.co.za> wrote:
>Christos Zoulas wrote:
>> In article <46A8BCB3.3060203@cat.co.za>, Ian McIntosh  <ianm@cat.co.za> wrote:
>>   
>>> Christos Zoulas wrote:
>>>     
>>>> Yes, compiling with MULTIPROCESSOR should work. Are your kernels compiled
>>>> with LOCKDEBUG or other options that might change the size of structures?
>>>>
>>>> christos
>>>>
>>>>   
>>>>       
>>> Thanks for the help, I cleaned my kernel and carefully config'ed it 
>>> again without LOCKDEBUG or any other options that change the structures 
>>> (as far as I can see). With this kernel I get somewhat better behaviour 
>>> in that PCI reads seem to work correctly. The problem now, is that once 
>>> the lkm loads it doesn't get any interrupts. A look at dmesg indicates 
>>> that my driver has been assigned ioapic0 pin 11 as its interrupt pin but 
>>> vmstat indicates there are no interrupts being generated for this pin.
>>> However the interrupt counts for ioapic0 pin 17 seem to be going wild. A 
>>> top on the system indicates that CPU0 is 72% idle when it was 100% idle 
>>> before the lkm loaded. Looking at /var/log/messages ioapic0 pin 17 is 
>>> assigned to uhci1 (USB controller I think).  Below is the vmstat before 
>>> and after the lkm load and the relevant dmesg.
>>>
>>>
>>> I would really appreciate any pointers here...
>>>
>>>     
>>
>> I don't remember the API for 3.x, but under current you need to both
>> map and establish the interrupt...
>>
>> christos
>>
>>   
>As far as I can see, for 3.x you have to do the same, I attach the code 
>fragment used below. I assume that if the interrupt mapping (and 
>establishing) was incorrect, the lkm would not function in a single 
>processor kernel either. In this case the lkm works perfectly for the 
>single processor kernel but with GENERIC.MPACPI it doesn't. Is there 
>anything else I can check to shed some more light?
>
>Ian
>
>  //  Set up interrupt handler.
>    if (pci_intr_map(pap,&intrhandle) != 0)
>    {
>      printf(": couldn't map PCI interrupt\n");
>      return;
>    }
>
>  intrstr = pci_intr_string(pap->pa_pc, intrhandle);
>
>  // choose TTY interrupt level
>  sc->sc_ih = pci_intr_establish(pap->pa_pc, intrhandle, 
>IPL_TTY,mov_intr, sc);
>  if(sc->sc_ih == NULL)
>    {
>      printf(": couldn't establish interrupt");
>      if(intrstr != NULL)
>        printf(" at %s", intrstr);
>      printf("\n");
>      return;
>    }
> 
>  if(intrstr != NULL)
>    printf("\r\n: intrstr %s dv_xname %s", intrstr, self->dv_xname);


The code looks correct. There must be some other bug in the interrupt
code. Please file a PR.

christos