Kernel Freeze with American Megatrends BIOS

[Roland Singer]

I tried these scenarios:

1. Booted the system without the bbswitch module. The nouveau module
   was loaded and is responsible for the power management of the GPU.
   The graphical session freezes after some minutes...

2. Booted the system without bbswitch and with nouveau blacklisted.
   Manually loaded bbswitch to switch off the discrete GPU.
   Same freeze after a while or by explicitly obtaining the GPU state.

Is there a possibility to switch off the discrete card without bbswitch?
If this is possible, then I could test this without nouveau and bbswitch
at all. If the system hangs, then it is not the video driver nor bbswitch.

Am 30.08.2016 um 16:08 schrieb Emil Velikov:
> On 30 August 2016 at 14:06, Bjorn Helgaas <helgaas at kernel.org> wrote:
>> On Tue, Aug 30, 2016 at 12:08:57PM +0200, Roland Singer wrote:
>>> Thanks for pointing it out.
>>>
>>> Yeah that's right. The system will hang randomly a few minutes later,
>>> because some certain actions in the graphical user session will trigger
>>> the freeze.
>>>
>>> I had a look at the function body of pci_read_config_dword:
>>>
>>>   #define PCI_OP_READ(size, type, len) \
>>>   int pci_bus_read_config_##size \
>>>       (struct pci_bus *bus, unsigned int devfn, int pos, type *value) \
>>>   {                                                                   \
>>>       int res;                                                        \
>>>       unsigned long flags;                                            \
>>>       u32 data = 0;                                                   \
>>>       if (PCI_##size##_BAD) return PCIBIOS_BAD_REGISTER_NUMBER;       \
>>>       raw_spin_lock_irqsave(&pci_lock, flags);                        \
>>>       res = bus->ops->read(bus, devfn, pos, len, &data);              \
>>>       *value = (type)data;                                            \
>>>       raw_spin_unlock_irqrestore(&pci_lock, flags);           \
>>>       return res;                                                     \
>>>   }
>>>
>>> I guess, that bus->ops->read(...) might be the trigger.
>>> Any hints how to continue debugging?
>>
>> It's not likely that the problem is in the bus->ops->read() path.  That
>> is used by every device driver, so a problem there would cause more
>> serious problems than what you're seeing.
>>
>> My guess would be some problem in the video driver or the bbswitch
>> thing.
>>
> FWIW I'm inclined to call it a bbswitch bug. It can (and does when
> needed) power off the dedicated GPU.
> 
> Depending on the platform different methods are used:
> 
> Sometimes the GPU driver will get 0xffffffff (or similar) when trying
> to read from the device mmio space. While one can say that the driver
> should attribute for this, IMHO it's a bad idea to have two drivers
> controlling the same hardware, let alone without any coordination
> between them.
> 
> IIRC in some cases the device can disappear from the PCI bus (not 100%
> sure this one). In which case a simple read can lead to a wide range
> of fireworks.
> 
> Disclaimer: it's been a while since I've looked into bbswitch so
> things might have changed/improved.
> 
> Regards,
> Emil
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>